VTT CREATES BUSINESS FROM TECHNOLOGY

Technology and market foresight • Strategic research • Product and service development • IPR and licensing
• Assessments, testing, inspection, certification • Technology and innovation management • Technology partnership
Dissertation VTT PUBLICATIONS 756

• • • VTT PUBLICATIONS 756

VTT PUBLICATIONS
738 Fusion Yearbook. Association EURATOM-TEKES. Annual Report 2009. Eds. by Seppo
Karttunen & Markus Airila. 2010. 136 p. + app. 13 p.
739 Satu Hilditch. Identification of the fungal catabolic D-galacturonate pathway. 2010.
74 p. + app. 38 p.
740 Mikko Pihlatie. Stability of Ni-YSZ composites for solid oxide fuel cells during reduction
and re-oxidation. 2010. 92 p. + app. 62 p.
741 Laxmana Rao Yetukuri. Bioinformatics approaches for the analysis of lipidomics

THIN FILM BULK ACOUSTIC WAVE DEVICES. PERFORMANCE OPTIMIZATION AND MODELING
data. 2010. 75 p. + app. 106 p.
742 Elina Mattila. Design and evaluation of a mobile phone diary for personal health
management. 2010. 83 p. + app. 48 p.
743 Jaakko Paasi & Pasi Valkokari (eds.). Elucidating the fuzzy front end – Experiences
from the INNORISK project. 2010. 161 p.
744 Marja Vilkman. Structural investigations and processing of electronically and protonically
conducting polymers. 2010. 62 p. + app. 27 p.
745 Juuso Olkkonen. Finite difference time domain studies on sub-wavelength aperture
structures. 2010. 76 p. + app. 52 p.
746 Jarkko Kuusijärvi. Interactive visualization of quality Variability at run-time. 2010.
111 p.
747 Eija Rintala. Effects of oxygen provision on the physiology of baker’s yeast Saccharomyces
cerevisiae. 2010. 82 p. + app. 93 p.
748 Virve Vidgren. Maltose and maltotriose transport into ale and lager brewer’s yeast
strains. 2010. 93 p. + app. 65 p.
749 Toni Ahonen, Markku Reunanen & Ville Ojanen (eds.). Customer value driven
service business development. Outcomes from the Fleet Asset Management Project.
2010. 43 p. + app. 92 p.
750 Tiina Apilo. A model for corporate renewal. Requirements for innovation management.
2010. 167 p. + app. 16 p.
751 Sakari Stenudd. Using machine learning in the adaptive control of a smart environment.
2010. 75 p. Tuomas Pensala
752 Evanthia Monogioudi. Enzymatic Cross-linking of ββ-casein and its impact on digestibility
and allergenicity. 2010. 85 p. + app. 66 p.
753 Jukka-Tapani Mäkinen. Concurrent engineering approach to plastic optics design.
2010. 99 p. + app. 98 p.
Thin Film Bulk Acoustic Wave
754 Sanni Voutilainen. Fungal thermostable cellobiohydrolases. Characterization and
protein engineering studies. 2010. 98 p. + app. 55 p.
Devices
756 Tuomas Pensala. Thin Film Bulk Acoustic Wave Devices. Performance Optimization
and Modeling. 2011. 97 p. + app. 73 p. Performance Optimization and Modeling

ISBN 978-951-38-7722-4 (soft back ed.) ISBN 978-951-38-7723-1 (URL: http://www.vtt.fi/publications/index.jsp)

ISSN 1235-0621 (soft back ed.) ISSN 1455-0849 (URL: http://www.vtt.fi/publications/index.jsp)
 VTT PUBLICATIONS 756

Thin Film Bulk Acoustic

Wave Devices
Performance Optimization and Modeling

Tuomas Pensala

Dissertation for the degree of Doctor of Science in Technology to be presented

with due permission of School of Science for public examination and debate
in Auditorium AS1 at Aalto University, School of Science (Otaniementie 17,
Espoo Finland) on 25th of February, 2011, at 12 noon.
ISBN 978-951-38-7722-4 (soft back ed.)
ISSN 1235-0621 (soft back ed.)
ISBN 978-951-38-7723-1 (URL: http://www.vtt.fi/publications/index.jsp)
ISSN 1455-0849 (URL: http://www.vtt.fi/publications/index.jsp)
Copyright © VTT 2011

JULKAISIJA – UTGIVARE – PUBLISHER

VTT, Vuorimiehentie 5, PL 1000, 02044 VTT

puh. vaihde 020 722 111, faksi 020 722 4374

VTT, Bergsmansvägen 5, PB 1000, 02044 VTT

tel. växel 020 722 111, fax 020 722 4374

VTT Technical Research Centre of Finland, Vuorimiehentie 5, P.O. Box 1000, FI-02044 VTT, Finland
phone internat. +358 20 722 111, fax + 358 20 722 4374

Technical editing Leena Ukskoski

Edita Prima Oy, Helsinki 2011

Tuomas Pensala. Thin Film Bulk Acoustic Wave Devices. Performance Optimization and Modeling
[Pietsosähköiset ohutkalvoresonaattorit. Suorituskyvyn optimointi ja mallinnus]. Espoo 2011. VTT
Publications 756. 97 p. + app. 73 p.

Keywords thin film, bulk acoustic wave, BAW, AlN, ZnO, resonator, filter, GHz, microacoustic,
modeling, interferometric imaging

Abstract
Thin film bulk acoustic wave (BAW) resonators and filters operating in the GHz
range are used in mobile phones for the most demanding filtering applications
and complement the surface acoustic wave (SAW) based filters. Their main ad-
vantages are small size and high performance at frequencies above 2 GHz. This
work concentrates on the characterization, performance optimization, and mod-
elling techniques of thin film BAW devices.
Laser interferometric vibration measurements together with plate wave disper-
sion modeling are used to extract the full set of elastic material parameters for
sputter deposited ZnO, demonstrating a method for obtaining material data
needed for accurate simulation of the devices.
The effectiveness of the acoustic interference reflector used to isolate the vi-
bration from the substrate is studied by 1-D modeling, 2-D finite element
method and by electrical and laser interferometric measurements. It is found that
the Q-value of reflector-based BAW resonators operating at 2 GHz is limited to
approximately 2000 by mechanisms other than leakage through the reflector.
Suppression of spurious resonances in ZnO resonators is studied in depth by
modeling and measurements. It is verified that the approximate mode orthogo-
nality is behind the suppression in boundary frame type ZnO devices operating
in the piston mode, but also another narrow band mode suppression mechanism
is found.
A plate wave dispersion based 2-D simulation scheme for laterally acousti-
cally coupled BAW resonator filters is developed and employed in designing of
experimental devices, which show both good agreement with the model predic-
tions and a remarkable 4.9 % relative bandwidth.

3
Tuomas Pensala. Thin Film Bulk Acoustic Wave Devices. Performance Optimization and Modeling
[Pietsosähköiset ohutkalvoresonaattorit. Suorituskyvyn optimointi ja mallinnus]. Espoo 2011. VTT
Publications 756. 97 s. + liitt. 73 s.

Avainsanat thin film, bulk acoustic wave, BAW, AlN, ZnO, resonator, filter, GHz, microacoustic,
modeling, interferometric imaging

Tiivistelmä
Pietsosähköisiin ohutkalvoihin perustuvia GHz-alueen mikroakustisia tila-
vuusaaltoresonaattoreita (BAW) ja niihin pohjautuvia suodattimia käytetään
matkapuhelinteknologiassa kaikkein vaativimmissa suodatuskohteissa. Teknolo-
gian keskeiset edut ovat pieni koko ja hyvä suorituskyky korkeilla ( 2 GHz)
taajuuksilla. Työssä keskityttiin BAW-resonaattoreiden karakterisointiin, suori-
tuskyvyn optimointiin ja mallinnustekniikkoihin.
Laserinterferometrimittauksia ja Lambin aaltojen dispersion mallinnusta käytetään
yhdessä ohutkalvomateriaalin elastisten materiaaliparametrien määrittämiseen.
Näin demonstroidaan menetelmä, jolla voidaan parantaa mm. mallinnuksen tarkkuutta
paremman materiaaliparametritiedon kautta.
Resonaattorit substraatista erottavan akustisen peilin ominaisuuksia analysoi-
daan 1D-malleilla, 2D-elementtimenetelmällä sekä kokeellisesti sähköisin ja
laserinterferometrimittauksin. Akustista peiliä parantamalla n. 2 GHz taajuudella
toimivien resonaattorien hyvyyslukua ei saada parannettua juurikaan yli 2000:n.
Sivuresonanssien vaimennusta reunapaksunnosmenetelmällä ZnO-resonaat-
toreissa analysoidaan mallinnuksella ja mittauksin. Tulokset osoittavat, että sivu-
resonanssit saadaan vaimennettua laajalla kaistalla niiden ominaismuotojen
heikkoon lineaariseen riippuvuuteen perustuen, mutta myös toinen kapeakaistainen
vaimennusefekti havaitaan.
Laatta-aaltojen dispersio-ominaisuuksiin perustuen kehitetään tehokas 2D-
mallinnusmenetelmä, jolla voidaan simuloida horisontaalisesti akustisesti kytkettyjä
suodattimia. Mallia vertaillaan kokeellisiin suodattimiin hyvin tuloksin, ja sen
avulla suunniteltu kokeellinen suodatin saavuttaa 4,9 % suhteellisen kaistan-
leveyden 2 GHz:n taajuudella.

4
Preface
Development of thin film Bulk Acoustic Wave resonators and filters was started
at VTT starting already in 1994 on the initiative of Nokia Mobile Phones, where
Juha Ellä and Helena Pohjonen became the principal driving forces. I had the
opportunity to join the highly talented team of VTT formed by Markku
Ylilammi, Jyrki Kaitila, Jyrki Molarius and Meeri Partanen, and lead by Ilkka
Suni, in 1998. I cannot overstate the value of that experience where I was intro-
duced to physics, modeling, characterization, and fabrication of the BAW reso-
nators and a highly innovative and focused way of working.
Great collaboration with the Materials Physics Laboratory of the Helsinki
University of Technology in the fields of finite element modelling and interfer-
ometric vibration imaging was carried with Tapani Makkonen, Jouni Knuuttila
and Prof. Martti M. Salomaa and others since the beginning of my work at VTT.
The fruitful co-operation has continued since then in many projects and outside
them with Kimmo Kokkonen, Olli Holmgren and others in the Optics and
Photonics group at Aalto University. Their optical measurement expertise pro-
vides a magnificent direct view into the operation of the devices. I wish to thank
Prof. M. Kaivola for supervising this work.
Most of the work presented in this thesis deals with optimization and deeper
study of the thin film BAW technology and was carried out in the forms of both
industrial co-operation and as academic collaboration. I want to thank the distin-
guished BAW team at Infineon Technologies for the excellent co-operation in
the optimization of the resonators and the BAW team at Philips Semiconductors
(later NXP Semiconductors) for very interesting work and fresh views, espe-
cially so in the field of acoustically coupled resonators.
In all of the work, the people, their knowledge, and infrastructure of the VTT
Microelectronics has played an instrumental part. Close collaboration with my
colleagues Johanna Meltaus, Tommi Riekkinen, James Dekker, Arto Nurmela,
Mervi Hämäläinen and others over the years has made possible this thesis. I
want to thank Ilkka Suni, Jouni Ahopelto and Hannu Kattelus for creating the
atmosphere for a successful work.
Finally, thanks to my family for support. I should cite my eldest son Ilma-
ri: ”Mikä ihmeen väitöskirja? Väitetäänkö siinä että aikuisten pitää leikkiä
enemmän lasten kanssa?”

5
Abstract
Thin film bulk acoustic wave (BAW) resonators and filters operating in the GHz
range are used in mobile phones for the most demanding filtering applications
and complement the surface acoustic wave (SAW) based filters. Their main ad-
vantages are small size and high performance at frequencies above 2 GHz. This
work concentrates on the characterization, performance optimization, and model-
ing techniques of thin film BAW devices.
Laser interferometric vibration measurements together with plate wave dis-
persion modeling are used to extract the full set of elastic material parameters for
sputter deposited ZnO, demonstrating a method for obtaining material data needed
for accurate simulation of the devices.
The effectiveness of the acoustic interference reflector used to isolate the vi-
bration from the substrate is studied by 1-D modeling, 2-D finite element method
and by electrical and laser interferometric measurements. It is found that the
Q-value of reflector-based BAW resonators operating at 2 GHz is limited to ap-
proximately 2000 by mechanisms other than leakage through the reflector.
Suppression of spurious resonances in ZnO resonators is studied in depth by
modeling and measurements. It is verified that the approximate mode orthogo-
nality is behind the suppression in boundary frame type ZnO devices operating in
the piston mode, but also another narrow band mode suppression mechanism is
found.
A plate wave dispersion based 2-D simulation scheme for laterally acousti-
cally coupled BAW resonator filters is developed and employed in designing of
experimental devices, which show both good agreement with the model predic-
tions and a remarkable 4.9 % relative bandwidth.

3
Tiivistelmä
Pietsosähköisiin ohutkalvoihin perustuvia GHz-alueen mikroakustisia tilavuusaal-
toresonaattoreita (BAW) ja niihin pohjautuvia suodattimia käytetään matkapuhe-
linteknologiassa kaikkein vaativimmissa suodatuskohteissa. Teknologian keskei-
set edut ovat pieni koko ja hyvä suorituskyky korkeilla (≥ 2 GHz) taajuuksilla.
Tässä työssä on keskitytty BAW-resonaattoreiden karakterisointiin, suorituskyvyn
optimointiin ja mallinnustekniikoihin.
Laserinterferometrimittauksia ja Lambin aaltojen dispersion mallinnusta käy-
tetään yhdessä ohutkalvomateriaalin elastisten materiaaliparametrien määrittämi-
seen. Näin demonstroidaan menetelmä, jolla voidaan parantaa parantaa mm.
mallinnuksen tarkkuutta paremman materiaaliparametritiedon kautta.
Resonaattorit substraatista erottavan akustisen peilin ominaisuuksia analysoi-
daan 1D-malleilla, 2D- elementtimenetelmällä sekä kokeellisesti sähköisin ja laser-
interferometrimittauksin. Todetaan, että akustista peiliä parantamalla ei n. 2 GHz
taajuudella toimivien resonaattorien hyvyyslukua saada parannettua juurikaan yli
2000:n.
Sivuresonanssien vaimennusta reunapaksunnosmenetelmällä ZnO-resonaatto-
reissa analysoidaan mallinnuksella ja mittauksin. Tulokset osoittavat, että sivures-
onanssit saadaan vaimennettua laajalla kaistalla niiden ominaismuotojen heikkoon
lineaariseen riippuvuuteen perustuen, mutta myös toinen kapeakaistainen vaimen-
nusefekti havaitaan.
Laatta-aaltojen dispersio-ominaisuuksiin perustuen kehitetään tehokas 2D-mal-
linnusmenetelmä, jolla voidaan simuloida horisontaalisesti akustisesti kytkettyjä
suodattimia. Mallia vertaillaan kokeellisiin suodattimiin hyvin tuloksin ja sen
avulla suunniteltu kokeellinen suodatin saavuttaa 4.9 % suhteellisen kaistanlevey-
den 2 GHz:n taajuudella.

4
Preface
Development of thin film Bulk Acoustic Wave resonators and filters was started
at VTT starting already in 1994 on the initiative of Nokia Mobile Phones, where
Juha Ellä and Helena Pohjonen became the principal driving forces. I had the
opportunity to join the highly talented team of VTT formed by Markku Ylilammi,
Jyrki Kaitila, Jyrki Molarius and Meeri Partanen, and lead by Ilkka Suni, in 1998.
I cannot overstate the value of that experience where I was introduced to physics,
modeling, characterization, and fabrication of the BAW resonators and a highly
innovative and focused way of working.
Great collaboration with the Materials Physics Laboratory of the Helsinki Uni-
versity of Technology in the fields of finite element modelling and interferometric
vibration imaging was carried with Tapani Makkonen, Jouni Knuuttila and Prof.
Martti M. Salomaa and others since the beginning of my work at VTT. The fruit-
ful co-operation has continued since then in many projects and outside them with
Kimmo Kokkonen, Olli Holmgren and others in the Optics and Photonics group
at Aalto University. Their optical measurement expertise provides a magnificent
direct view into the operation of the devices. I wish to thank Prof. M. Kaivola for
supervising this work.
Most of the work presented in this thesis deals with optimization and deeper
study of the thin film BAW technology and was carried out in the forms of both
industrial co-operation and as academic collaboration. I want to thank the distin-
guished BAW team at Infineon Technologies for the excellent co-operation in the
optimization of the resonators and the BAW team at Philips Semiconductors (later
NXP Semiconductors) for very interesting work and fresh views, especially so in
the field of acoustically coupled resonators.
In all of the work, the people, their knowledge, and infrastructure of the VTT
Microelectronics has played an instrumental part. Close collaboration with my
colleagues Johanna Meltaus, Tommi Riekkinen, James Dekker, Arto Nurmela,
Mervi Hämäläinen and others over the years has made possible this thesis. I want
to thank Ilkka Suni, Jouni Ahopelto and Hannu Kattelus for creating the atmo-
sphere for a successful work.
Finally, thanks to my family for support. I should cite my oldest son Ilmari:
"Mikä ihmeen väitöskirja? Väitetäänkö siinä että aikuisten pitää leikkiä enemmän
lasten kanssa?"

5
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Tiivistelmä . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
List of papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Author’s Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1 Introduction 10
1.1 The Mobile Phone and Filtering . . . . . . . . . . . . . . . . . . 10
1.2 Microacoustic Filters for Radio Frequencies . . . . . . . . . . . . 12

2 BAW Basics 14
2.1 The Crystal Resonator . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 The Equivalent Circuit Representation . . . . . . . . . . . . . . . 18
2.3 Electrical Characterization of BAW Resonators . . . . . . . . . . 22
2.4 Definitions for Effective Coupling Coefficients . . . . . . . . . . 23

3 One Dimensional Picture of the BAW Resonator 27

3.1 One Dimensional Acoustic Model of the Resonator . . . . . . . . 27
3.2 Reflector Design by Transmission Line Analogy . . . . . . . . . . 30

4 BAW Filters and Resonators 38

4.1 BAW Ladder Filters . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2 Response of a Real BAW Resonator and Filter . . . . . . . . . . . 40

5 Lateral Effects in Thin Film BAW Resonators 45

5.1 Plate Waves and Dispersion . . . . . . . . . . . . . . . . . . . . . 45
5.2 Lateral Energy Trapping . . . . . . . . . . . . . . . . . . . . . . 50
5.3 2-D Modeling of BAW Resonators . . . . . . . . . . . . . . . . . 53
5.4 FEM Modeling of Thin Film BAW Devices . . . . . . . . . . . . 57
5.5 Spurious Resonance Suppression . . . . . . . . . . . . . . . . . . 63
5.6 Laterally Acoustically Coupled Resonator Filters . . . . . . . . . 70

6 Interferometric Imaging of BAW Resonators 76

7 Loss Mechanisms in a BAW Resonator 80

7.1 Experimental Determination of the Reflector Performance . . . . 80

8 Discussion 85

9 Summary 88

6
References 90

Appendices
Papers 1–6 98

Papers 1–6 are not included in the PDF version.

Please order the printed version to get the complete publication
(http://www.vtt.fi/publications/index.jsp).

7
 L IST OF PAPERS

1. T. Makkonen, T. Pensala, J. Vartiainen, J.V. Knuuttila, J. Kaitila, and M.M.

Salomaa, “Estimating Materials Parameters in Thin-Film BAW Resonators
Using Measured Dispersion Curves”, IEEE Trans. on Ultrasonics, Ferro-
electrics and Frequency Control, vol. 51, 2004, pp. 42–51.

2. T. Pensala, R. Thalhammer, J. Dekker, and J. Kaitila, “Experimental Inves-

tigation of Acoustic Substrate Losses in 1850-MHz Thin Film BAW Res-
onator”, IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control,
vol. 56, 2009, pp. 2544–2552.

3. K. Kokkonen, T. Pensala, and M. Kaivola, “Dispersion and Mirror Trans-

mission Characteristics of Bulk Acoustic Wave Resonator ”, IEEE Trans.
on Ultrasonics, Ferroelectrics and Frequency Control, vol. 58, 2011, pp.
215–225.

4. T. Pensala and M. Ylilammi, “Spurious Resonance Suppression in Gigahertz-

Range ZnO Thin-Film Bulk Acoustic Wave Resonators by the Boundary
Frame Method: Modeling and Experiment”, IEEE Trans. on Ultrasonics,
Ferroelectrics and Frequency Control, vol. 56, 2009, pp. 1731–1744.

5. T. Pensala, J. Meltaus, K. Kokkonen and M. Ylilammi, “2-D Modeling of

Laterally Acoustically Coupled Thin Film Bulk Acoustic Wave Resonator
Filters”, IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control,
vol. 57, 2010, pp. 2537–2549.

6. J. Meltaus, T. Pensala, K. Kokkonen and A. Jansman, “Laterally Coupled

Solidly Mounted BAW Resonators at 1900 MHz”, Proc. IEEE Ultrasonics
Symposium, 2009, pp. 847–850.

8
Author’s Contribution
The work was carried out at VTT Technical Research Centre of Finland during
the years 2002–2010. Experiments of Paper 1 were planned by the author together
with T. Makkonen and J. Knuuttila (Materials Physics Laboratory, Helsinki Uni-
versity of Technology). The author performed the analysis of the experimental
samples apart from the interferometric measurements and wrote the correspond-
ing parts of the paper. The study of Paper 2 was done in collaboration with Infi-
neon Technologies GmbH. The author planned the experiment with the co-authors
and performed the characterization and analysis of the samples and results and
wrote the paper. The author planned the experiments of Paper 3 together with K.
Kokkonen (Applied Physics, Helsinki University of Technology) and performed
the simulations and analysis of the results. For Paper 4 the author performed the
measurements, simulations and analysis and wrote the paper, while the samples
were results of earlier collaboration by VTT and Nokia Oyj. The model of Paper
5 was developed by the author by extending on the earlier work of M. Ylilammi
(VTT) while the experiments were planned and conducted in co-operation with J.
Meltaus (VTT) and K. Kokkonen (Applied Physics, Helsinki University of Tech-
nology). In Paper 6 the author contributed to the design and modeling of the
experimental devices.

9
1 Introduction
Thin film bulk acoustic wave (BAW) resonators and filters form one of the en-
abling technologies for the radio technology used in the mobile phones. They
serve as GHz-range passband filters selecting the appropriate range of frequen-
cies for the mobile phone radio to receive and transmit its communications signals,
blocking out the other unwanted frequencies.

1.1 The Mobile Phone and Filtering

The mobile communications revolution started in the early 1990’s, bringing a
portable radio – the mobile phone – that could be used as an ordinary phone but
practically anywhere into the reach of anyone in a matter of less than a decade.
While the number of produced mobile phones increased at incredible yearly rates,
the size and price of the devices were reduced in equally astonishing pace. This
created a demand for new kinds of ultra-miniature, high performance, mass-pro-
ducible and cheap components making the rapid development of the mobile phones
possible.
A mobile phone uses certain defined frequency ranges to transmit and receive
signals. For example, an early "900 MHz" Enhanced-GSM phone transmits in
channels within the range 880–915 MHz and receives in the range 925–960 MHz
[1]. When receiving, the mobile phone has to be able to pick up a weak sig-
nal from a base station possibly several kilometers away at the presence of other
nearby phones and other kinds of radios transmitting powerful signals. At the
air interface, the signal is picked up by the antenna, and must then be amplified
by a component called the low-noise-amplifier, the LNA, for further processing.
On the other hand, when transmitting, the mobile phone has to output a signal
of significant power (even several watts) in order to reach the base station. This
signal is produced by the power amplifier, known as the PA. The LNA is designed
to work with very weak signals and it is critical to block interfering unwanted
signals from entering in it by filtering. On the other hand, the PA is designed to
blast out a powerful radio signal, and care must be taken to limit the signal to the
allocated frequency range with the aid of band-pass filters. In a GSM phone, the
reception and transmission are sequenced by division into successive time slots
by a switch alternating between the receive and transmit paths. In spread spec-
trum systems such as the CDMA (for Code Division Multiple Access) and the
WCDMA (Wideband-CDMA), the reception and transmission occur simultane-
ously without division into time slots in frequency bands typically very close to
each other. Obviously, there is a need for very efficient isolation of the receive
and transmit signals. This is realized by a high performance duplexer, a three port
filter having the antenna at one port and the receive and transmit paths at the two
remaining ports.
10
 Figure 1: Illustration of usage of pass-band filters in the front-end
of a GSM phone (top) and a CDMA phone (bottom). RX stands for
receive path and TX for transmit path.

In the very simplest example, one filter is used for the receive path and another
for the transmit path to select the desired frequencies and to suppress the undesired
ones. The usage of band-pass filters in a GSM and in a CDMA phone radio front
end is illustrated in Fig. 1 by simplified schematics. In actual mobile phones,
more filters are needed but the detailed architectures of mobile radio systems are
outside the scope of this work.
The development of the mobile phones has not only made them smaller and
cheaper, but has increased their internal complexity vastly. A typical high end mo-
bile phone can communicate in several GSM bands and several WCDMA bands.
As an example, the Nokia N8 smartphone works in the GSM bands at 850, 900,
1800 and 1900 MHz and in WCDMA bands at 850, 900, 1700, 1900, and 2100
MHz [2]. In addition, the same device has WLAN, Bluetooth, and FM radios. A
significant number of high performance filters are needed in extremely small form
factor to facilitate realization of such a communications device.

11
1.2 Microacoustic Filters for Radio Frequencies
Filtering a radio signal requires structures that are comparable to the wavelength
of the carrier wave. At 1 GHz, the wavelength of an electromagnetic wave in air is
30 cm and at 2 GHz half of that, both being larger than the dimensions of a mobile
phone. The key to the extreme miniaturization of GHz range filters is transforma-
tion of the electromagnetic wave to an acoustic wave via the piezoelectric effect.
The velocities of acoustic waves in solids are typically in the range of 3000–11000
m/s, which is five decades less than the velocity of an electromagnetic wave. The
wavelength is reduced by the same fraction – into the micrometer range in case
of GHz frequencies. This allows for fabrication of microacoustic filters for the
GHz-range with sub-millimeter dimensions.
Practically all the filters in a mobile phone are nowadays of the microacoustic
type. The dominating technology utilizes surface acoustic waves (SAW) travel-
ing along the surface of a single crystal piezoelectric substrate such as LiTaO3 ,
LiNbO3 or quartz [3]. The thin film BAW filters utilizing a longitudinal wave in
the thickness direction of a vacuum-deposited polycrystalline piezoelectric thin
film have been developed to extend the range of applicability of microacoustic
filters.
The SAW filter technology is well applicable for frequencies up to 2 GHz,
but above that the critical dimensions of the devices become so small that prob-
lems with economical manufacturing, robustness and power handling arise. Also,
SAW filters suffer from fairly large temperature drift of frequency. Another draw-
back of the SAW devices is that the Q-values, translating directly to the steepness
of passband skirts, remain below 1000. It is in the high frequency end, that the
BAW filter technology provides advantages in Q-values, temperature stability, and
power handling capability. In the low frequencies the BAW filters have as disad-
vantage a fairly large size. For a detailed comparison of SAW and BAW filter
technologies, see [3].
The break-through application band of the thin film BAW technology was
the US-PCS CDMA utilized in North America. The transmit band is at 1850–
1910 MHz and the receive band at 1930–1990 MHz, leaving only 20 MHz (1.05
%) gap in between [1]. The filtering used to be done with a ceramic duplexer
which was unimaginably large to be fit into a mobile phone of today. The SAW
technology had great difficulties trying to bring about a microacoustic solution due
to the low Q-values and large temperature drift leading to specification violations
in the guard band between the receive and transmit bands. The thin film BAW
technology, instead, managed to provide a duplexer with high performance and
greatly reduced size. This was allowed mainly by the high Q and low temperature
drift of the AlN resonators of Hewlett Packard Laboratories (later Agilent and
Avago Technologies) [4]. Later, SAW duplexers for the same application were

12
 Figure 2: Photograph of a 932 MHz ZnO resonator based on a
SiO2 − W acoustic reflector on glass substrate. The active area
(grainy square at top right) has an area of 300 × 300 µm2 .

developed [5], but the BAW filters have since established a position in the high
performance and high frequency end of mobile phone filters, the US-PCS and
some WCDMA bands being the main applications [3].
In this work, mainly the properties of a single BAW resonator (Fig. 2), which
is a basic building block of a ladder filter ([6], Fig. 16), are considered. It is the
characteristics of the individual resonators that dictate the achievable performance
of the ladder filters. The field of ladder filter design itself is not entered but filters
based on lateral acoustical coupling of adjacent resonators are discussed.

13
2 BAW Basics
2.1 The Crystal Resonator
A bulk acoustic wave crystal resonator consists of a thin slab of piezoelectric
crystal which is metallized on both sides for applying an exciting voltage (Fig. 3).
The plate exhibits a resonance whenever an integer multiple of half an acoustic
wavelengths fits within the resonator thickness t, at frequencies
vN
fN = , (1)
2t
where v is the acoustic velocity of the wave mode in question (shear or longi-
tudinal), N an integer multiplier, and t the thickness of the crystal. Only odd
values of N lead to piezoelectric coupling as even N correspond to antisymmetric
modes. The electrical frequency response of the resonator is that of a capacitor
except near a resonance, where it has in addition to a resonance (at fr in Fig.
4) an anti-resonance (fa ), the frequency separation of which is a measure of the
strength of the piezoelectric effect in the device. Resonators with very low losses,
i.e. high Q-value, can be fabricated from for example quartz, yielding for highly
frequency selective resonators. The traditional method of fabricating crystal os-
cillators include fabrication of bulk crystals, from which thin plates are diced,
ground and polished down to a desired thickness. It is the dimensions produced
by this manufacturing process that determine the resonance frequency of the res-
onators, followed by a possible trimming step. It is difficult to thin down a crystal
plate down to the micrometer scale, limiting the quartz resonators’ frequencies
typically in the range below 200 MHz [7].
The crystal resonators can be connected to an active circuit to form an oscilla-
tor, or they can be connected in a ladder topology to form a passive filter (see Sec.
4.1). The most common material used for crystal resonators, quartz, exhibits only
very weak piezoelectricity allowing for narrow band filters only [8].
The thin film BAW resonator technology is based on the same principal struc-
ture, but the piezoelectric material as well as the electrodes, are produced by vac-
uum deposition of thin films. Standard lithography and etching methods used in
microelectronic circuit fabrication are employed to pattern the films into devices.
This allows for micrometer-range dimensions and subsequently GHz-range reso-
nance frequencies. Furthermore, the piezoelectric thin films such as ZnO and AlN
exhibit a much stronger piezoelectric effect than quartz, yielding for a possibility
to create passband filters with a relative bandwidth of several percent.
Thin film technology requires a substrate such as silicon or glass wafer to
provide a mechanical support for the films. The vibration, however, needs to be
contained within the thin film resonator structures. Two methods for providing

14
 Figure 3: The structure of a crystal resonator. Stress distribution of the
fundamental half-wavelength thickness resonance is illustrated with
the dashed line.

this are used: 1) In membrane type resonators (often called FBAR) the substrate
or a sacrificial layer between the resonator and the substrate is removed to generate
an air interface at the lower surface of the resonator [9, 10]. 2) In Bragg reflec-
tor based resonators, suggested by Newell already in 1965 [11], often referred to
as the solidly mounted resonators (SMR), an acoustic interference reflector con-
sisting of alternating layers of high and low acoustic impedance each quarter of
an acoustic wavelength in thickness, is fabricated between the resonator and the
substrate. Both of these types are in industrial production [12, 13, 14]. The main
variations of the air gap resonators are illustrated in the top row and the SMR type
at the bottom of Fig. 5.
Aluminium nitride is the piezoelectric material used in all commercial appli-
cations of thin film BAW, although a fair amount of research has been conducted
with ZnO-based technology [15, 16, 17, 18, 19]. The key properties of the two
materials are compared in Table 1. The electromechanical coupling coefficient is
slightly stronger in ZnO, which actually exhibits the strongest piezoeffect of all
non-ferroelectric materials. This very desirable property for filter applications is
somewhat eaten up by the twice as large temperature drift of frequency as com-
pared to AlN. The dielectric permittivities of the two materials are very similar,
close to 10 · ε0 . AlN exhibits very large sound velocities, leading to thicker layers
for a given frequency. The acoustic impedances are of the same order, but the
Poisson ratio differs significantly being very large for ZnO and fairly small for
AlN.
The main reasons for AlN being the dominant material in the industry is that it

15
 Figure 4: Magnitude of impedance of a piezoelectric crystal resonator
showing resonance-antiresonance behavior.

Figure 5: An air gap or FBAR type resonator structure fabricated by

sacrificial layer etching (top left), a self supported membrane res-
onator fabricated by through-wafer etching (top right), and a Bragg
reflector/SMR type resonator (bottom).

is compatible with integrated circuit (IC) fabrication processes unlike ZnO. This
is important not because of the monolithic integration possibility of thin film BAW
and active silicon circuits as visioned earlier [9], but because an IC facility can be
used to fabricate AlN BAW devices removing the need for a dedicated facility
[22]. Another key issue is the small temperature drift of AlN as compared to
ZnO and especially as compared to SAW devices. AlN and ZnO are most often
deposited by sputtering. In order to display piezoelectric effect of the same order
as in bulk crystal, the piezofilms must be deposited under very well controlled
conditions. A considerable effort in the development of such deposition processes
had to take place before commercial manufacturing of thin film BAW devices was
feasible.
16
 Table 1: Comparison of ZnO and AlN properties relevant to wave
propagation along the c-axis (aligned with z-axis) [20, 6, 21]. The
e2
definition K 2 = ε3333cE is used for the electromechanical coupling co-
33
efficient.

Material AlN ZnO

density ρ (kg/m3 ) 3405.5 5680
longitudinal wave velocity vl (m/s) 11250 6366
shear wave velocity vs (m/s) 6329 2734
long. wave impedance Zl (kg/m2 /s · 106 ) 38.3 36.2
shear wave impedance Zs (kg/m2 /s · 106 ) 21.6 15.5
Poisson ratio ν 0.246 0.387
piezoelectric stress coefficient e33 (C/m2 ) 1.5 1.32
dielectric permittivity ε33 (ε0 ) 9.66 10.2
electromechanical coupling coefficient K 2 (%) 6.5 9.15
temperature coefficient of frequency TCF (ppm/K) -25 -60
CMOS compatible yes no
crystal symmetry 6mm (hex.) 6mm (hex.)

Figure 6: Structures of a two different solidly mounted ZnO res-

onators fabricated at VTT. The ZnO resonator (left) is based on glass
substrate and mostly wet etching, while the AlN resonator one on the
right uses a high-resistivity Si substrate and dry etching processes.
Both employ a W − SiO2 reflector and a boundary frame type spu-
rious resonance suppression structure [23]. A mass loading layer for
frequency shifting is shown on the ZnO resonator only but can be
applied in the same manner also for the AlN resonators.

Detailed cross sectional structures of two different actual SMR BAW res-
onators fabricated at VTT are shown in Fig. 6. On the left is shown a ZnO
resonator [23] of the kind studied in Papers 4 and 3, and on the right an AlN
resonator.

17
 Figure 7: The modified Butterworth-van Dyke (mBvD) equivalent cir-
cuit representation of a piezoelectric resonator. The modifications
consists of adding the loss resistors Rs , R0 and Rp to the standard
BvD circuit. Additional series inductor Ls is also often used in series
with Rs .

2.2 The Equivalent Circuit Representation

Piezoelectric crystal and thin film resonators can be described with the Butter-
worth van Dyke (BvD) equivalent circuit [24]. The circuit, slightly modified from
what is usually understood as the BvD circuit is shown in Fig. 7. The modifica-
tions by Larson et al. [25] include the addition of two loss resistors (Rs , R0 ) to
improve the match to real thin film BAW devices. A series inductance Ls is added
to represent the inductance formed by the electrode layout of the resonator and
the finite resistivity of the piezolayer can be described by the resistor Rp , which
is usually insignificant.
In its standard form, the BvD circuit consists of a mechanical resonance branch
(Lm , Cm , Rm ) and a capacitance C0 in parallel. The capacitance C0 is simply the
plate capacitance formed by the piezoelectric film or crystal sandwiched between
the electrodes and is given by (ignoring fringing field contribution)
AεS
C0 = , (2)
t
where A is the area of the resonator, εS is the permittivity of the piezolayer in
constant strain, and t is the thickness of the piezolayer. Nominally, the series
resistor Rs describes the resistance of the electrodes, the resistor R0 the dielectric
losses in the piezo-material, Rm the acoustic losses (viscous, radiative), and Rp
the finite resistivity of the piezolayer.
Such a circuit exhibits a resonance-antiresonance behavior. In resonance, the
reactance of the mechanical branch,
1
Xm = ωLm + , (3)
ωCm

18
vanishes and current is drawn by the circuit through the mechanical branch, lim-
ited only by resistances Rs and Rm . The resonance frequency ωr is thus obtained
from
1
ωr2 = . (4)
Lm Cm
In anti-resonance, the reactance around the loop formed by the mechanical branch
and the C0 -branch,
1 1
X0 + Xm = + ωLm + , (5)
ωC0 ωCm
vanishes, leading to an anti-resonance frequency at

Lm (C0 + Cm )
ωa2 = . (6)
C0 Cm
In the anti-resonance, very little current is drawn by the circuit, but a resonant
current circulates around the loop formed by the plate capacitance C0 and the me-
chanical branch. Electrically, the resonance and anti-resonance can be considered
as the short circuit or series resonance and the open circuit or parallel resonance,
respectively.
While the expression (6) is not exact, it is a very good approximation sufficient
for all practical purposes. The exact resonance and antiresonance frequencies can
be solved from the minimum and maximum, respectively, of the mBvD-circuit
impedance
1
ZmBvD = jωLs + Rs + 1 1 . (7)
R0 + jωC0
+ 1
Rm + jωC +jωLm
m

The relative difference of the resonance and antiresonance frequencies is an

important figure of merit related to the effective coupling coefficient of the res-
onator. Using the definition
ω2 − ω2
2
keff = a 2 r, (8)
ωa
[26], one finds the direct connection to the equivalent circuit parameters

2 Cm
keff = . (9)
C0 + Cm
It is to be noted that several different definitions for the coupling coefficient are
being used in the literature (see Section 2.4 for a brief discussion).

19
 The loss resistors of the circuit define the Q-values. A commonly used defini-
tion for the Q-value of an electrical circuit is
¯ ¯
ω ¯ dφ ¯
¯ ¯
Q= ¯ ¯
¯ dω ¯
, (10)
2 ω0

where φ = arg(Zin ) is the phase of the input impedance (7) of the circuit. For the
mBvD circuit, which exhibits a resonance - anti-resonance (i.e. series and parallel
resonance) behavior, this definition leads to the Q-values
ωr Lm
Qr ≈ (11)
Rs + Rm
and
ωa L m
Qa ≈ (12)
R0 + Rm
at the resonance ωr and at the antiresonance ωa , respectively. It is seen that the Qr
is loaded by the electrical series resistance, whereas Qa is defined by the internal
losses only. This is important in measurement of actual devices since it implicates
that the Q-value at the antiresonance is not affected by the electrode resistances
nor by a possible contact resistance in the measurement. Further, one can define a
purely mechanical Q-value as
ωa Lm
Qmech. = . (13)
Rm
If one measures the complex input impedance Z(ω) of the resonator, one can
extract the resonance frequencies ωr and ωa from e.g. the minimum and maxi-
mum of |Z(ω)| and calculate the Q-values (11) and (12). The plate capacitance
C0 can be either calculated based on the known geometry of the resonator or also
extracted from a measurement. With this information one is able to fix the equiv-
alent circuit parameters using the following associations:

2
keff
Cm = 2
C0 , (14)
1 − keff
1
Lm = 2
, (15)
ωr Cm
ωr Lm
Rm = . (16)
Qr
An actual resonator can exhibit more than a single mechanical resonance, such
as higher harmonic resonances and spurious resonances near the main resonances

20
 Figure 8: The many-mode modified Butterworth-van Dyke (mBvD)
equivalent circuit representation of a piezoelectric resonator. Each
mechanical branch corresponds to a single resonance.

[27]. The nature of these will be discussed later. Additional mechanical branches
can be added in parallel to the the equivalent circuit to represent each mode [28].
The admittance of the multi-resonance circuit (Fig. 8) is (assuming for sim-
plicity that Rs and R0 are small enough to be neglected)
N
X 1
Y = jωC0 + 1 , (17)
p=1 Rmp + jωLmp + jωCmp

where the p indexes the resonance modes and ωp refers to the resonance frequency
of each mode.
With the connections (14)–(16) this can be expressed as
N
X kp2 1
Y = jωC0 + CQω
2 0 p p Qp (ωp2 −ω 2 )
(18)
p=1 1 − kp 1+ jωωp

emphasizing the fact that the parameter sets (C0 , keff,p , Qp , ωp ) suffice to define the
response of the device. This expression is of importance in developing the models
used for spurious resonances and lateral inter-resonator coupling in Papers 4 and
5.
The equivalent circuit elements can be obtained by 1) fitting the circuit re-
sponse to a measured response of a resonator or 2) by using an acoustic model to
simulate a resonator and derive its electrical response from the results. The for-
mer is typically used to reduce the measured frequency response of a resonator to
a limited set of parameters that can be used to assess the resonator performance
and to further calculate how a filter comprising such resonators would perform.
The latter serves for the basic resonator design tasks.

21
2.3 Electrical Characterization of BAW Resonators
The thin film BAW resonators are typically characterized by measuring the scat-
tering parameters Sij of a resonator with a vector network analyzer (VNA) on
wafer level with RF-probes. A resonator can either be fabricated as a one- or two-
port device (Fig. 9 a) and b), respectively). To assess the performance of a single
resonator it suffices to measure only the reflection coefficient S11 of a one-port
configuration, but filters are obviously two-port devices.
The VNA is calibrated to the plane of the probe tips, indicated by the dashed
line in Fig. 9. The device under test consists of everything beyond that plane. The
remaining parasitic contributions from the measurement setup that are difficult to
remove by any calibration include the contact resistance from the probes to the
contact pads and the inductance of the loop formed by the electrodes from tip to
another.
The VNA measurement is performed within the characteristic impedance Z0 =
50 Ω environment. From the measured S11 , the input impedance is obtained by
1 + S11
Zin = Z0 . (19)
1 − S11
One can directly compare this with the impedance of the mBvD equivalent circuit
(7). Typically, a fit of the mBvD-equivalent circuit is performed to the measure-
ment data in order to reduce it into a small set of descriptive parameters. In order
to be able to make comparisons between resonators, the fitting procedure must
be very stable even in the presence of noise, variation in contact resistance, and
spurious resonances. In Paper 2, the difficulties in the fitting and extraction of
the Q-value were discussed and it was concluded that in order to analyze small
differences in the performance of resonators, direct comparison of raw data is also
needed.
Within the same study, an equivalent circuit fitting procedure was constructed
by improving the one developed earlier by Infineon Technologies GmbH. The
main improvement achieved was the increased stability of the fitted Q-value in
the presence of spurious ripple in the measurement data. The fitting algorithm is
summarized below (for the definitions of the fitting parameters, see Section 2.2):
1. Make a first rough guess for the circuit parameters based on directly ex-
tracted ωr , ωa , C0 , Qr , and Qa .
2. Fit C0 and Qmech using S11 data points with a large radius on the Smith chart
(to avoid bias from spurious ripple).
3. Fit Rs and R0 using input impedance phase data outside resonance with
an asymmetric cost function that prevents from fitting too large an Rs (Rp
defaults to 1 MΩ unless detected to have a significantly small value).

22
 Figure 9: One-port (a) and two-port (b) configurations for electrical
measurements of thin film BAW resonators. The RF-probes are rep-
resented by the gray triangles, S standing for signal and G for ground
connection. The dashed line represents the plane of calibration.

4. Fit again C0 and Qmech using data points with a large radius on the Smith
chart.

5. Refit ωr and ωa using data points with large radius on the Smith chart.

It is important to use a fitting procedure instead of direct extraction by equa-

tions (8), (10) etc. as noise and spurious resonances can cause large local de-
viations in the directly extracted ωr , ωa , and the derivative of the phase of the
impedance not reflecting the actual overall performance of the device [29].

2.4 Definitions for Effective Coupling Coefficients

In the literature, several different definitions are being used for the effective elec-
tromechanical coupling coefficient of piezoelectric resonators (for a summary, see
[29]). One of the most often used is
" Ã !#−1
2π fs π fs
K = tan , (20)
2 fp 2 fp
where fs and fp are the series and parallel resonance frequencies, respectively.
They stand for the resonance and anti-resonance frequencies and from here on the
serial and parallel resonances are used synonymously with resonance and antires-
onance, respectively.

23
 Definition (20) has it origins in an analytical 1-D solution for a piezoelectric
slab with infinitesimally thin electrodes. Approximations of this include

π 2 fs fp − fs
K2 = , (21)
4 fp fp

and
π 2 fp − fs
K2 = . (22)
4 fp
Another version of the latter with fp in the denominator replaced by fs has also
been used. The usage of several different definitions, reporting squared or non-
squared values, makes comparison of reported resonator performances quite diffi-
cult. It is better to always report the exact resonance frequencies fs and fp along
with an explicitly defined coupling coefficient in order to avoid misunderstand-
ings.
The fundamental phenomenon to be measured is the effective electromechani-
cal coupling coefficient which manifests itself in the relative frequency difference
of fs and fp . This coefficient is in thin film BAW resonators affected by material
parameters (piezoelectric film quality), electrodes, the rest of the layer stack (es-
pecially in the SMR type devices), lateral design, size of the resonator [30], and
parasitic capacitances and inductances.
We have chosen to use the definition
v
u 2
uf − f2
keff = t p 2 s , (23)
fp

from [26] due to its simplicity as a function of the resonance frequencies, and even
more importantly, due to its direct connection to the equivalent circuit representa-
tion via (9):
2 Cm
keff = . (24)
C0 + Cm
The fundamental meaning of the electromechanical coupling coefficient is to
measure the efficiency of the piezoelectric device in converting electrical energy
to mechanical energy and vice versa. This is clear in the definition of Berlincourt
[31] for the coupling,
U2
k2 = m (25)
Ue Ud
where Um is the so-called mutual energy (energy capable of conversion), Ue is
the elastic energy, and Ud is the dielectric energy. Following Kaitila [23, 32], we
assume that only vertically (z) directed stress and electric fields exist, but that the

24
fields have a lateral (x)-dependence of a separable form. The energy terms can be
expressed as:
1Z
Um = dzz Tz Ez dv, (26)
2v
1Z E 2
Ue = s T dv, (27)
2 v zz z
and
1Z T 2
Ud = ε E dv. (28)
2 v zz z
Here dzz is the piezoelectric strain constant for longitudinal z-directed strain, Tz is
the longitudinal strain in the z-direction, Ez is the electric field in the z-direction,
εTzz is the corresponding permittivity, and sE zz is the compliance for longitudinal
strain in the z-direction. The integration is performed over the volume of the
device.
Now, following Kaitila [32], assume that in a vertical resonator with one lateral
dimension (x) the stress field can be expressed as

Tz = Tz0 (z)u(x), (29)

and the electric field as

Ez = Ez0 (z)E0 (x), (30)
where u(x) is a lateral shape function (mode shape) and E0 (x) is the lateral dis-
tribution of the electric field (e.g. unity over electrode and zero elsewhere) and
the primed variables are the z-dependent fields. Substitution of (29)–(30) into
(26)–(28) leads to
Z Z
dzz
Um = Tz0 (z)Ez0 (z) dz u(x)E0 (x) dx (31)
2 z x
Z Z
sE
zz 0
Ue = Tz 2 (z) dz u(x)2 dx (32)
2 z x

εTzz Z 0 2 Z
Ud = Ez (z) dv E02 (x) dx. (33)
2 z x

The Berlincourt formula (25) now gives

· ¸2 · ¸2
R R
d2zz Tz0 (z)Ez0 (z) dz E0 (x)u(x) dx
k2 = · R z0 2 R 0 · R x2 R . (34)
E T
szz εzz 2
Tz (z) dz Ez (z) dz u (x) dx E02 (z) dx
z z x x

25
 2
The first two terms form the 1-D-coupling (or the Mason-model coupling) k1D ,
solely dependent on the vertical fields and the last term establishes the dependence
on the lateral distribution of the fields,
· ¸2
R
E0 (x)u(x) dx
x
k 2 = k1D
2
R R . (35)
u2 (x) dx E02 (z) dx
x x

This result is utilized in the modeling of the lateral effects in the BAW devices in
Papers 4 and 5. It is to be noted that as the mutual energy (26) is an integral over
the product of stress and electric field, no contribution to the coupling coefficient
comes from areas where the electric field is zero. The important practical conse-
quence of this is that vibration in areas where no electrical field is applied reduces
the effective piezoelectric coupling of the resonator.

26
3 One Dimensional Picture of the BAW Resonator
3.1 One Dimensional Acoustic Model of the Resonator
The equivalent circuit of a BAW resonator does not contain any direct informa-
tion of its operation physics if the circuit elements are not associated to the actual
physical device. The work horse for the modeling of the BAW resonator is an
electro-acoustic 1-D model applied in the wave propagation direction discarding
the lateral dimensions. The model of Mason is often used for this purpose [33, 24].
It presents the acoustic layers as acoustic transmission lines and the piezoelectric
layer with an additional electrical port via a transformer. Instead of the standard
Mason model, a transfer matrix model formulated by M. Ylilammi [34] is used in
this work. The transfer matrix model is equivalent to the Mason model but em-
phasizes the acoustics not resorting to electrical circuit description. As it serves to
review the basic physics of piezoelectric resonators, the formulation of the transfer
matrix method is outlined below.
Consider a stack of layers indexed by i, each of thickness ti (Fig. 10). It is
assumed that in each of the layers, there are two plane waves traveling up and
down, with complex amplitudes Ai and Bi , respectively:

ui (z) = Ai e−jβi z ejωt + Bi ejβi z ejωt . (36)

Here βi = vac.,i /ω is the wave number in layer i, vac.,i being the acoustic wave

Figure 10: Geometry of the 1-D transfer matrix model for a multilay-
ered BAW resonator.

27
velocity in that layer. The stress in the layers can be found by using the constitutive
relations
Ti = c E
i Si − ei Ei (37)
Di = εSi Ei + ei Si , (38)
where Ti is the stress, Di the electric displacement field, cE i the stiffness in a
constant electric field, ei is the piezoelectric stress coefficient, Si = ∂ui /∂z is the
strain, and Ei is the electric field. From (38) one can solve the electric field as Ei
ei Ji
Ei = − S −j S ,
S i
(39)
εi εi ω
where the connection Ji = jωDi between the current density and the electric
displacement field Di has been used assuming harmonic time dependence of the
fields. Substituting this into (38) one obtains
∂ui hi Ji
Ti = c i +j , (40)
∂z ω
where ci = cE 2 S S
i + ei /εi is the piezoelectrically stiffened stiffness, hi = ei /εi is the
piezoelectric coefficient, and Ji the current density trough the layer (excitation).
For a non-piezoelectric layer with ei = 0 the stress is simply
∂ui
Ti = cE
i. (41)
∂z
The boundary conditions at the layer interfaces require that that the displace-
ment and the stress are continuous
ui+1 (zi+1 = 0) = ui (zi = ti ) (42)
Ti+1 (zi+1 = 0) = Ti (zi = ti ). (43)
Layer-wise z-coordinates extending from zero to the thickness of the layer ti have
been employed here.
Applying (42)–(43) and solving for the amplitudes Ai+1 , Bi+1 as function of
those in the layer below Ai , Bi , one finds a layer transfer matrix Mi and a source
term Ci to yield " # " #
Ai+1 Ai
= Mi + Ci . (44)
Bi+1 Bi
The 2 × 2 matrix Mi is a function of the acoustic impedances Zi of the layer mate-
rials, layer thicknesses and frequency/wave number Mi = Mi (Zi , Zi+1 , ki , ω, ti ),
in detail  ³ ´ ³ ´ 
1  e−jβi ti ³1 + ZZi+1
i
´
ejβi ti 1 − ZZi+1
³
i
´ .
Mi = (45)
2 e−jβi ti 1 − ZZi ejβi ti 1 + ZZi
i+1 i+1

28
The source term Ci = Ci (hi,i−1 , Ji,i−1 , ω, Zi ) contains the piezoelectric contribu-
tion and is of the from
" #
hi Ji − hi−1 Ji−1 1
Ci = , (46)
2ω 2 Zi −1

but is calculated recursively as is evident from (44).

Possible boundary conditions at the extreme ends of the stack can be 1) vac-
uum interface, where the stress T must vanish or 2) infinite half-space of some
material, where only outgoing waves exist (assuming that the excitation can only
be done by driving current density Jpiezo through the piezolayer).
For given boundary conditions, the amplitudes (Ai , Bi ) can be calculated in
each layer i of the 1-D stack as a response to a harmonic current density Jpiezo
present at the piezolayer. The vibration response of the stack, ui (z), is obtained
via (36), the strain from Si = ∂ui /∂z, and the stress from (40). The voltage over
the piezolayer, Vpiezo , is then obtained by integrating (39) over the thickness of the
piezolayer tpiezo leading to

Ji t i ei h −jβi ti jβi ti
i
Vi = −j + S
A i (1 − e ) + B i (1 − e ) , (47)
εS ω εi
where index i refers to the piezolayer.
Equation 39 yields the electrical impedance Zin of the device as the input
impedance of the resonator is obtained from
Vpiezo
Zin = , (48)
AJpiezo

where A is the assumed area of the resonator. This scaling into two lateral dimen-
sions is needed in order to compare the 1-D predictions with measurements. Here
we have assumed constant current drive, but having obtained the input impedance,
it is straightforward to use (48) to calculate the current density due to constant
voltage drive and use that as excitation in a simulation again to obtain the vibra-
tion response to a constant voltage drive.
In the lossless case, the wavenumber βi is given by
ω
βi = . (49)
vi
Viscous losses can be incorporated in the model by employing a complex wavenum-
ber β̂ and acoustic impedance Ẑ [24],

ω ηi ω 2
β̂i = −i 3 (50)
vi 2vi ρi

29
and
ci ηi ω
Ẑi = −i . (51)
vi 2vi
Here ηi is the viscosity and ρi the density of the material of layer i.
With the 1-D model one can do the basic design of the resonators, such as to
find the layer thicknesses that yield the desired resonance frequency, to optimize
the layer stack for effective coupling coefficient, estimate the reflector-limited Q-
value, etc. Practically all BAW resonator design starts with the 1-D model.
The 1-D model can be applied for longitudinal as well shear waves by just
using the material parameters for the appropriate wave mode.
In Fig. 11, the at-resonance displacement, stress, and strain fields in a 932
MHz ZnO resonator with a W − SiO2 reflector are shown. Such a resonator
structure was studied in Papers 3 and 4. The half-wave resonance contained in
the electrodes and the piezolayer (three leftmost layers) and the suppression of
the amplitude towards the substrate (right) is evident.

3.2 Reflector Design by Transmission Line Analogy

In the reflector based resonators, it is instrumental to optimize the reflector struc-
ture in order to fabricate high performance devices. The simplest way of designing
the acoustic Bragg reflector for a longitudinal mode resonator is to set the thick-
ness of each reflector layer to be of a quarter of an acoustic wavelength, i.e.,
vL
tλ/4 = , (52)
4f0
where vL is the longitudinal bulk acoustic wave velocity and f0 is the targeted
operation frequency. This provides maximal reflection for a pure plane wave of
frequency f0 arriving at the reflector. To find out what the coefficient of reflec-
tion and bandwidth of operation of the reflector actually are, one can utilize an
analogy to electrical transmission lines [11, 35]. One can use the expressions for
electrical transmission lines for calculating the reflection and transmission of ver-
tically propagating bulk acoustic plane waves in a 1-D layer stack simply just by
replacing the electrical wave propagation parameters (impedance, wave velocity)
by the acoustical ones.
A single section of transmission line of length t, acoustic impedance Zc and
sound propagation velocity v, terminated to an acoustic impedance Zb is presented
in Fig. 12. The impedance Za seen by an acoustic plane wave arriving from
infinite half-space on the left is

Zb cos(βt) + jZc sin(βt)

Za = Zc , (53)
jZb sin(βt) + Zc cos(βt)

30
 −11
x 10 displacement amplitude ±|u|
5

|u| (m) 0
Au
air Al ZnO SiO2 W SiO2 W substrate
−5
0 1 2 3 4 5 6 7 8 9
−6
x 10
7
x 10 stress amplitude ±|T|
1
T (Pa)

−1
0 1 2 3 4 5 6 7 8 9
−6
x 10
−5
x 10 strain amplitude ±|S|
5

0
S

−5
0 1 2 3 4 5 6 7 8 9
distance from top (m) −6
x 10

Figure 11: Displacement, stress, and strain amplitudes on a 932 MHz

ZnO SMR calculated by the 1-D model at the series resonance fre-
quency fs . Air interface is on the left and substrate on the right. The
layers are separated by vertical lines. The three leftmost layers are the
top electrode, the piezolayer, and the bottom electrode, respectively.

Figure 12: Transmission line principle for calculating the properties

of an acoustic Bragg reflector.

where the wave number is given by

2πf
β= . (54)
v
Here Zb may be a constant acoustic impedance if the termination is to an infi-
nite half-space (or a thick substrate with a rough backside in practice), or to a

31
frequency dependent impedance of another transmission line structure. Equation
(53) can be used iteratively to obtain the impedance seen by a plane wave inci-
dent upon a series of an arbitrary number of transmission line sections (i.e. thin
film layers) by using the Za obtained in the previous step as a new terminating
impedance Zb . Having calculated the impedance Za for a layer stack, one can
then calculate a reflection coefficient for a bulk plane wave arriving from an infi-
nite half space having an acoustic impedance Zref from
Za − Zref
r= . (55)
Za + Zref
In the process of designing an acoustic Bragg reflector, the impedance Zref can
be chosen to be either the impedance of the bottom electrode, or if the bottom
electrode itself is considered to be a part of the reflector itself, the impedance of
the piezolayer. It is to be noted that the same procedure allows calculating the
transmission characteristics for longitudinal and shear waves by just choosing the
appropriate propagation parameters (acoustic impedance and wave velocity). In
[36] Marksteiner et al. discovered that the shear reflection characteristics of the
Bragg reflector can indeed have profound effects on the Q-value of a longitudinal
mode resonator. They also suggest inspecting a logarithmic transmittance of the
form
TdB = 10 log10 (1 − |r|2 ) (56)
instead of the reflection coefficient (55) to resolve small differences important for
high-Q resonators. This practice is adopted here.

Comparison of Reflector Layer Stacks

In Table 2, the longitudinal wave acoustic impedance ratios of some typical mate-
rials used in the Bragg reflectors of thin film BAW devices are listed. Remarkably,
W and SiO2 provide an impedance ratio of almost 8. With Mo and SiO2 one can
reach a ratio of 5 while the ones using dielectric materials (AlN, ZnO, Ta2 O5 ) as
the high-Z material offer ratios of 3 and below.
Transmittances of several different quarter wavelength reflector structures de-
signed for 1 GHz operation frequency are shown in Fig. 13. The material pairs
are implicated in the plot legends multiplied by a number indicating the number of
layer pairs, 2.5 × W − SiO2 for example standing for the layer sequence substrate
- SiO2 - W -SiO2 - W - SiO2 - resonator. The longitudinal wave transmittances
are shown in the upper part of Fig. 13 where one can observe that the impedance
ratio determines the bandwidth of operation of the reflector, 2.5 × W − SiO2 re-
flector being by far the widest. The shear wave transmittances are shown in the
lower part of the plot, showing that most reflectors have a very high transmittance

32
 Table 2: Acoustic impedance ratios of material pairs used in SMR
reflectors.

Materials ZL -ratio
W − SiO2 7.9
Mo − SiO2 5.0
AlN − SiO2 3.0
ZnO − SiO2 2.8
Ta2 O5 − SiO2 2.4

for shear waves at the operation frequency. The ZnO − SiO2 reflector forms an
exception – significant shear reflectance is provided. The wide shear reflection
band of the 2.5 × W − SiO2 however reaches close to the operation frequency,
yielding for possible manipulation of the shear transmission characteristics by
slightly deviating from the quarter-wavelength thicknesses [36].
The minimum transmittances at f0 , defining the reflector-limited Q-values for
the longitudinal wave show the fact that 2.5 × W − SiO2 , 4.5 × AlN − SiO2 or
4.5 × ZnO − SiO2 reflectors can provide roughly equal minimum transmittance
at the operation frequency. The 3.5 × Mo − SiO2 reflector provides a clearly
lower minimum transmittance and the 4.5 × Ta2 O5 -SiO2 reflector is inferior to
the others. In the shear transmittance plot (lower part of Fig. 13) the ZnO − SiO2
stands out with its unique shear isolation property.
The minimum transmittance and thus the reflector-limited Q-value is deter-
mined by the number of layer pairs for a given impedance ratio. With a material
selection having a lower impedance ratio, one has to use a larger number of layers
to obtain a minimum transmittance equal to that provided by materials of higher
impedance ratio. When using metals as the high-Z layers, the number must be
minimized as they need to be patterned in order to prevent capacitive shunting
along the metal layers. When using dielectrics only, no patterning is needed and
it may be economical to deposit even a large number of layers. As will be dis-
cussed later, the transmittance of the reflector, however, does not alone determine
the resonator performance.
The Q-value of a resonator is defined by [28]
Etot.,stored
Q = ω0 , (57)
Ploss,ave.

where Etot.,stored is the total energy stored in the resonator and Ploss,ave. is the
average loss power. If the losses are for example only in the form of energy
transmitted to the substrate described by transmittance T , the Q-value takes the

33
 0
long. transm. (dB)
−10 3.5 x Mo−SiO
2
−20 2.5 x W−SiO2
4.5 x AlN−SiO2
−30
4.5 x Ta2O5−SiO2
−40 4.5 x ZnO−SiO2
−50
200 400 600 800 1000 1200 1400 1600 1800
f (MHz)

0
shear transm. (dB)

−10

−20

−30

−40
200 400 600 800 1000 1200 1400 1600 1800
f (MHz)

Figure 13: Transmittances (56) for longitudinal (top) and shear waves
(bottom) for longitudinal mode quarter-wavelength reflectors of dif-
ferent materials. The 2.5 × W − SiO2 reflector provides by far the
widest bandwidth due to highest impedance ratio, the next widest be-
ing the 3.5 × Mo − SiO2 reflector. The all-dielectric reflectors have
clearly smaller bandwidths due to smaller impedance ratio. The
AlN − SiO2 and ZnO − SiO2 reflectors with 3.5 pairs reach the same
minimum transmittance as the 2.5 × W − SiO2 reflector, while the
3.5 × Mo − SiO2 is superior and 4.5 × Ta2 O5 -SiO2 would require a
layer pair more to reach the same level. For the shear waves, all but
the 3.5 × ZnO − SiO2 reflector show high transmissivity. However,
the shear band edge of the 2.5 × W − SiO2 reflector is near the oper-
ation frequency.

Table 3: Comparison of 2.5 × W − SiO2 , 5.5 × Ta2 O5 − SiO2 and

5.5 × Pt − Ta2 O5 − SiO2 reflectors.

stack keff Q
W − SiO2 0.2233 5500
Ta2 O5 -SiO2 0.2159 5130
Pt − Ta2 O5 -SiO2 0.2228 6250

34
 0
long. transm. (dB)
−10

−20
2.5 x W−SiO2
−30
5.5 x Ta2O5−SiO2
−40
5.5 x Pt−Ta2O5−SiO2
−50
200 400 600 800 1000 1200 1400 1600 1800
f (MHz)

0
shear transm. (dB)

−10

−20

−30

−40
200 400 600 800 1000 1200 1400 1600 1800
f (MHz)

Figure 14: Calculated transmittances of 2.5 × W − SiO2 ,

5.5 × Ta2 O5 − SiO2 and 5.5 × Pt − Ta2 O5 − SiO2 reflectors.
The bandwidth of the 5.5 × Ta2 O5 − SiO2 reflector can be seen
to be narrower than that of the 2.5 × W − SiO2 reflector, but re-
placing just the topmost Ta2 O5 by Pt significantly widens it. The
minimum transmittance of the first two are comparable, while the
5.5 × Pt − Ta2 O5 − SiO2 provides over 10 dB smaller transmittance
and significantly reduces shear transmittance.

form
Etot.,stored ω0
Q = ω0 = . (58)
T · Etot.,stored T
This way one can translate the reflector transmittance into a Q-value. From the
nature of the Q-value being inversely proportional to relative losses, it follows that
1 X 1
= , (59)
Qtot i Qi

where the summation is over different loss mechanisms. Now, inspecting a single
loss mechanism such as the reflector transmittance, we can interpret the resulting
Q-values as the ones that the device would exhibit in case all other loss mech-
anisms where much smaller in magnitude. We can thus interpret the minimum
transmittance resulting from the reflector simulations as the inverse of a ’reflector
limited Q-value’.

35
 ELECTRODES REFLECTOR
10
PIEZO
2.5 x W−SiO2
|T| (MPa)

5.5 x Ta2O5−SiO2
5
5.5 x Pt−Ta2O5−SiO2

0
0 5 10 15 20
z (µm) (distance from surface)

0.03

0.02
|u| (nm)

0.01

0
0 5 10 15 20
z (µm)

Figure 15: Comparison of acoustic fields at fs for resonators utilizing

a metal 2.5 × W − SiO2 , dielectric 5.5 × Ta2 O5 − SiO2 , and a com-
posite 5.5 × Pt − Ta2 O5 − SiO2 reflector. The upper plot shows the
stress distribution through the stack while the bottom plot shows the
displacement. The all dielectric 5.5 × Ta2 O5 − SiO2 reflector shows
a much larger vibration outside the resonator (indicated by dashed
lines) leading to reduced keff . This is in large part remedied by replac-
ing the topmost layer of Ta2 O5 by Pt.

Hybrid Pt − Ta2 O5 − SiO2 Reflector

As an example differing from the mainstream, a reflector combining both dielec-
tric and metal high-Z layers is discussed next.
The BAW technology of Philips Semiconductors (later NXP Semiconductors)
used to be based on Ta2 O5 − SiO2 all-dielectric reflector motivated by the econ-
omy of depositing all the layers in the same vacuum process and avoiding expen-
sive patterning steps [37]. Also, a fully planar substrate is automatically provided
for the subsequent resonator fabrication steps. However, neither the Q-value nor
the coupling coefficient of the devices were sufficient for the aimed filter applica-
tion.
In Fig. 14 it is shown that the all-dielectric 5.5 × Ta2 O5 − SiO2 reflector pro-
vides a minimum transmittance comparable to the 2.5 × W − SiO2 metal solution.
However, the effect of transmittance on Q is not the only role that the acoustic

36
reflector has on the electrical performance of the resonator. As is already evident
from the fact that more layer pairs are needed for lower acoustic impedance ratio
materials, the vibration penetrates deeper into the reflector before being extin-
guished. This has a direct effect on the effective electromechanical coupling keff
of the resonator. As concluded in Section 2.4, all the vibration energy outside the
piezolayer sandwiched between the electrodes cannot contribute to the keff and
thus reduce it. Thus, for maximizing the coupling coefficient of the resonator it is
crucial to provide the reflection of the acoustic energy as near the resonator itself
as possible.
Instead of completely switching to metallic high-Z layers, only the top-most
high-Z Ta2 O5 can be replaced by platinum, a material already being used in the
process for the bottom electrode, to form a hybrid metal-dielectric reflector in-
volving only a single additional lithography step. Platinum as a high-impedance
material provides an impedance ratio of 5.3 with SiO2 . The performance im-
provement is clear in the prediction by the 1-D model shown in Table 3, where
the W − SiO2 reflector provides a benchmark. Both the coupling coefficient and
reflector-limited longitudinal wave Q-value are improved to values comparable
with the W − SiO2 reflector. In order to facilitate comparison, simulations have
been carried out with identical hypothetical resonator structures on top of the var-
ied reflectors – symmetric 400 nm thick Mo electrodes and a 3.28 µm AlN piezo-
layer.
In the transmittance plots of Fig. 14, it can be observed that the hybrid
Pt − Ta2 O5 − SiO2 reflector provides a clearly wider bandwidth than the all-
dielectric reflector for longitudinal waves and that there is significantly reduced
transmittance for the shear waves, too. The minimum transmittance is actually
improved from the all-dielectric case and also as compared to the W − SiO2
benchmark. Results of a one-dimensional simulation of the vibration amplitude
(stress and displacement) in the layer stacks are shown in Fig. 15, where the rea-
son for the reduced coupling coefficient of the all-dielectric solution stands out as
higher vibration amplitude in the reflector, outside the volume enclosed between
the electrodes.
The BAW team at NXP Semiconductors further optimized this structure for
the shear reflecting properties, the dispersion characteristics, and for the suppres-
sion of spurious resonances with the boundary frame method [23] – the work
resulted in high performance spurious resonance free BAW resonators exhibiting
a Q-value of over 2000 [38].

37
4 BAW Filters and Resonators
4.1 BAW Ladder Filters
The most commonly used method to build band pass filters from BAW resonators
is to use two kinds of resonators shifted to slightly different frequencies (labeled
HI and LO for higher and lower frequency resonators, respectively, in the illustra-
tions) and of slightly different area (capacitance), and to connect them electrically
in the ladder topology (Fig. 16). Simulated impedances of such individual res-
onators are shown in the upper part of Fig. 17, and the transmission (S21 ) of a
3 21 -stage filter (Fig. 16 c) comprising of such resonators in the lower part. A
photograph of an actual ZnO-based 3-stage ladder filter fabricated at VTT for the
E-GSM receive band is shown in Fig. 18.
The electrical performance of the individual resonators dictate the filter per-
formance that can be obtained by the ladder filter constructed from them. Most
importantly, the effective coupling coefficient of the resonators defines the obtain-
able flat passband width, as illustrated by simulation in Fig. 19. The resonator
coupling coefficients represent a range from AlN quality obtainable from a sub-
optimal sputtering system and conditions (k eff = 0.19) to an industrial level op-
timized process (k eff = 0.23). The filter labeled ’overtuned’ shows an attempt
to generate a passband equal to that obtained with high coupling coefficient res-
onators with lower keff resonators by just increasing the frequency shifting be-
tween the higher and lower frequency resonators. The flatness of the passband is
lost and a dent in the passband is generated. This demonstrates the fundamental
limit set by the resonator coupling coefficient bandwidth of the ladder-type BAW
filter.
The effect of the Q-value on the filter performance shows in a slightly more
subtle way in the shape of the passband. It affects the minimum attenuation of
the pass-band and the sharpness of the pass-band corners. The latter is often more
critical for fulfilling industrial specifications. A simulation of this effect done by
varying Q while keeping keff fixed is shown in Fig. 20.
Of the performance parameters, the coupling coefficient and the Q-value are
of the highest importance for filter applications. A third very important perfor-
mance criterium for the resonator is the amount of ripple present in the frequency
response, a matter discussed in detail in the next sections. It is the optimization
of the resonator performance for these three aspects that made possible the appli-
cation of BAW technology in mobile phones and is also the main focus of this
work.

38
Figure 16: Topology of a ladder filter. The resonators with label HI
are the series resonators having a higher frequency and the resonators
with the label LO have a lower frequency. a) A single ladder stage
forming the basic building block of such filters. b) a 3-stage ladder
filter. c) 3 21 -stage ladder filter.

4
10 2
Z arg(Z) (rad)
s
1
|Z| (Ω)

Z
2 p
10 0

−1
0
10
900 950 1000 1050 −2
1100
f (MHz)
0
S21 (dB)

−50

−100

900 950 1000 1050 1100

f (MHz)

Figure 17: Impedances of individual resonators of a ladder filter (top)

and S21 of a 3 21 -stage ladder filter comprising of them. A slight over-
tuning,i.e. offset between the anti-resonance of the low frequency res-
onators and the resonance of the high frequency resonators, has been
applied.

39
 Figure 18: A photograph of a ZnO-based thin film BAW 3-stage lad-
der filter. The filter is aimed for the E-GSM receive band at 925-
960 MHz. The upper row of resonators are the series resonators (HI)
and the lower row resonators are the shunt resonators (LO) having a
slightly lower center frequency. The width of the filter is roughly 1.5
mm.

−5

−10
S21 (dB)

−15
keff=0.19

−20 keff=0.21
k =0.23
eff
−25 k =0.21 overtuned
eff

−30
960 970 980 990 1000 1010 1020 1030
f (MHz)

Figure 19: Calculated effect of keff for filter passband of a 3.5 stage
ladder filter. A frequency shift of 1.2 × (fp − fs ) is applied for the
LO resonators except the for the ’overtuned’ case, where an attempt
to compensate for the non-optimal keff has been done by a larger fre-
quency shift resulting in a dent at the passband.)

4.2 Response of a Real BAW Resonator and Filter

In Figures 21–22 the measured input impedance of an actual ZnO BAW resonator
and the scattering parameter S11 (reflection coefficient) are shown together with a
40
 0

−5 Q=500
Q=800
S21 (dB)

Q=1500
Q=2000

−10

−15
970 980 990 1000 1010 1020
f (MHz)

Figure 20: Calculated effect of resonator Q-value for the filter pass-
band of a 3 21 -stage ladder filter. The increase of the Q-value can
be seen to reduce the minimum passband attenuation and make the
passband shape more sharp-cornered, which is often most critical for
meeting specifications.

1-D simulation result and a fit of the mBvD equivalent circuit to the measurement.
The most striking difference of the experimental results to those simulated by the
1-D model and the equivalent circuit fit is the ripple above fs with a period of a
few MHz. Visible at the bottom near the capacitive 50 Ω point in the Smith chart
plot of Fig. 22 are also additional ripples forming small circles.
The magnitude of the impedance (upper plot in Fig. 21) is quite well predicted
by the 1-D model, except for the fact that the minimum and maximum values are
exaggerated. These are a consequence of unrealistically high Q-values, i.e., low
losses in the 1-D model. In the phase plot the same effect is seen as steeper phase
transitions at fs and fp . Also to be noted is the fact that the phase transitions are
equally steep as predicted by the 1-D model, indicating equal Q-values at fs and
fp . In the measurement, Q at fs is lower due to a significant series resistance
absent in the 1-D model.
The mBVD circuit fitted to the measurement shows a good match to the mag-
nitude of Zin . The spurious undulations are almost within the linewidth of the plot.
However, in the phase of Zin the unability of the single resonance mBVD-circuit
to follow the spurious response is clearer. In the Smith chart plot of Fig. 22 it can
be seen that the 1-D model that overestimates the Q-values forms an unrealisti-
cally large circle, while the mBVD fit neatly encircles the measured curve so that

41
the spurious dips deviate only towards the center of the Smith chart.
There is one type of ripple present in the SMR type resonators that the 1-D
model can explain: the so called backside reflections which result from standing
wave resonances within the full thickness of the substrate. This effect is discussed
in Paper 2.
The other type of ripple present in both SMR and FBAR type devices re-
sponsible for that present in Figs. 21–22 is typically a stronger effect. There the
frequency spacing is not constant, depends on the the lateral shape of the res-
onators, and decreases with increasing lateral dimensions of the resonator. This
ripple is caused by laterally standing wave resonances – a phenomenon associ-
ated with lateral propagation of plate wave or Lamb wave modes [39, 40, 23, 41].
Understanding the origin and being able to suppress these lateral spurious modes
has been one of the major issues in the successful development of thin film BAW
resonator filters for commercial use in mobile phones. The origin of the spurious
modes, a method for their suppression, and simulation techniques are discussed in
Section 5 and in Paper 4. To underline the importance of effective suppression of
the spurious modes, Fig. 23 shows the measured transmission characteristics of
two filters, one being constructed of resonators with ripple free responses and the
other one suffering from significant spurious resonances.
A measured wide band response from extending from 50 MHz to 5 GHz is
also presented in Fig. 24 (thick line). A total of 10 resonances are observed,
two below (very close to each other) and 7 above the fundamental resonance. All
of these are thickness resonances in nature that are readily predicted by the 1-
D model (dashed line). There is some deviation in the predicted and observed
resonances, especially towards the higher frequencies. This is due to inaccuracies
in the thicknesses and material parameters of the layers. A method for extracting
elastic material parameters for the thin films is presented in Paper 1.
The discrepancy in the background levels of |Zin | and arg(Zin ) is due to a
series inductance present in the actual devices.

42
 MEASUREMENT
1D MODEL
|Z| (Ω) mBVD FIT

0
10
920 930 940 950 960 970
f (MHz)
2
arg(Z)

−2
920 930 940 950 960 970
f (MHz)

Figure 21: Measurement, 1-D model prediction, and equivalent circuit

fit of the input impedance (magnitude and phase) of a ZnO BAW res-
onator. The measurement shows ripple caused by the spurious lateral
resonances and reduced Q-values, especially so at fs , as compared to
the 1-D model.

MEASUREMENT
1D MODEL
mBVD FIT

Figure 22: Measurement, 1-D model prediction, and equivalent cir-

cuit fit of the reflection coefficient S11 of a ZnO BAW resonator on
the Smith chart. Measurement shows the undulation caused by the
spurious resonances while the 1-D model predicts a clean circle, the
radius of which is larger than that measured due to higher Q. The
equivalent circuit fit encloses the measurement tightly.

43
 0

−5 FILTER 1
FILTER 2
S21 (dB)

−10

−15

−20
940 950 960 970 980 990
f (MHz)

Figure 23: Measured transmissions (S21 ) of two 3-stage ladder filters

such as the one shown in Fig. 18. Filter 1 (thin line) has strong spu-
rious resonances and a low Q-value, whereas Filter 2 (thick line) has
weak spurious resonances and a high Q-value.

MEASUREMENT
1D MODEL
|Z| (Ω)

0
10
1000 2000 3000 4000 5000
f (MHz)
2
arg(Z)

−2
1000 2000 3000 4000 5000
f (MHz)

Figure 24: Wideband measurement and 1-D simulation of the input

impedance of a ZnO BAW resonator showing a multitude of thick-
ness resonances in addition to the main resonance at 930 MHz. The
1-D model (dashed line) predicts all resonances observed in the mea-
surement, but with some deviation in actual frequencies.

44
5 Lateral Effects in Thin Film BAW Resonators
Ideally, one would like a BAW resonator to be practically one-dimensional in its
operation and use the lateral dimensions only to achieve proper matching to the
required impedance level and to tailor the stop band rejection in ladder filters. The
practical devices have a width to thickness ratio on the order of 100, but still the
finite lateral dimensions and the mechanical boundary conditions at the edges do
have profound effects on the resonator performance. First, the vibration needs
to be trapped within the resonator in the lateral dimensions as well as in the ver-
tical direction. Otherwise a high-Q resonance is not supported. Secondly, the
ripple caused by lateral resonances needs to be suppressed in order to obtain clean
frequency response. Both of these issues are connected to the fact that acoustic
vibration excited in the thickness direction of the resonator can propagate later-
ally as plate waves. The understanding and controlling of the lateral propagation
characteristics of the plate waves, i.e. plate wave dispersion, plays a key role in
the design of BAW resonators.

5.1 Plate Waves and Dispersion

A thin film bulk acoustic wave resonator consists essentially of a stack of plates
with different acoustic properties. Propagation of sound waves in such structures
has been studied in detail already since the 19th century in connections ranging
from vibrations of the crust of the earth to industrial non-destructive ultrasonic
inspection of sheet materials.
In the bulk, three different vibration modes can propagate: a longitudinal mode
and two differently polarized shear modes. The longitudinal mode is character-
ized by compression and elongation in the direction of the propagation, whereas
the shear modes consist of motion perpendicular to the direction of propagation
with no local change of volume. The propagation characteristics of the three bulk
modes depend on the material properties and propagation direction respective to
the crystal axis orientations. For a discussion of bulk propagation characteristics,
see for example [21] and [24]. In the following, we limit the discussion to one
longitudinal and one shear mode for simplicity.
When a propagating bulk wave encounters an interface with another material
with different material properties, transmitted and reflected waves are generated.
In such an event, a pure bulk mode is reflected as a pure bulk mode of the same
type only in exact normal incidence [28]. Generally, the stress and particle conti-
nuity conditions at the interface require an incident wave to be reflected and trans-
mitted as a combination of all bulk modes propagating in different angles. The
phenomenon is called mode conversion and is governed by Snell’s Law, which

45
 Figure 25: Illustration of the general case of acoustic reflection and
transmission/refraction for an interface of two isotropic materials. A
longitudinal wave LI arrives at the interface with an angle of incidence
of θL1 giving rise to a reflected longitudinal wave LR with an angle of
reflection of equal magnitude θL1 . Additionally, a reflected shear wave
SR is generated with a different angle of reflection θS1 . Transmitted
waves of longitudinal and shear type LT , ST are generated with angles
of refraction θL2 and θS2 . The lengths of the arrows represent the wave
number. The continuity conditions require that the component of the
wave vector along the interface must be equal for all five waves.

for an interface of two isotropic acoustic materials is

ω ω ω ω
sin θL1 = sin θS1 = sin θL2 = sin θS2 , (60)
vL1 vS1 vL2 vS2
where ω is the angular frequency of the wave, vL1 , vS1 , vL2 , vS2 are the wave veloc-
ities for longitudinal and shear waves in material 1 and in material 2, respectively
[28]. The angles θL1,S1,L2,S2 are illustrated in Fig. 25. In the case of anisotropic
materials, Snell’s law takes a more complicated form.
The discussion is limited now to two dimensional structures having one ver-
tical dimension (z) and a single horizontal dimension (x). In a structure formed
of a finite number of stacked plates (Fig. 26), pure bulk modes can propagate
and form standing wave resonances in the (z)-direction perpendicular to the layer

46
 Figure 26: Laterally propagating plate waves in a multilayer stack are
formed by such combinations of partial bulk shear and longitudinal
waves that can fulfil the continuity conditions at the layer interfaces
and boundary conditions at the top and bottom of the stack.

interfaces. If, however, the wave propagation direction of any of the bulk wave
components, defined by the wave vector β, ~ deviates from perpendicular, mode
conversion according to Snell’s law takes place at every interface. It turns out
that with certain combinations of propagation directions, repeating mode conver-
sion at reflections at the interfaces reproduces the original field in a periodical
manner in the lateral x-direction [28]. This kind of wave modes propagating in
the x-direction are called plate wave modes or Lamb waves, some basic types of
which are illustrated in Fig. 27 for a single plate. The laterally finite BAW res-
onator operation is actually defined by the characteristics of these waves as such
a resonator practically never can really resonate in a pure bulk wave mode. The
presence of laterally propagating waves in vertically stacked layer structures has
the important consequence that the operation of the intended vertical-mode res-
onators becomes sensitive to the lateral boundary conditions and geometry of the
device as reflections of these waves from lateral discontinuities will occur.
It is of crucial importance to be able to simulate and measure the lateral propa-
gation characteristics, i.e. the plate wave dispersion relations, in the BAW devices

47
 Figure 27: Illustration of some plate wave modes for a single plate:
the 1st thickness extensional (TE1 ), the 2nd thickness shear (TS2 ), the
1st thickness shear (TS1 ), and flexural modes, from top to bottom,
respectively. Two full lateral wavelengths are shown for each mode.
The frequency increases from bottom to top with the exception that
the order of TE1 and TS2 depends on the Poisson ratio of the plate
material.

in order to be able to design high performance devices. The dispersion of the plate
wave modes can be calculated for example by matrix methods, some of which are
reviewed by Lowe in [42]. In Fig. 28, the dispersion curves calculated by the
Thomson-Haskell matrix method (numerical implementation by the author) for
a single plate of ZnO are shown. The dispersion curves are presented as a plot
of frequency against the horizontal component βx of the wave vector of the bulk
waves, as it is an invariant of the system – at a given frequency all bulk wave com-
ponents at all layers in solid contact must have the same βx . This is illustrated in
Fig. 26 by the fact that while the wave vectors of partial bulk waves have differ-

48
 Figure 28: An example of dispersion curves calculated with the
Thomson-Haskell matrix method for a single 3 µm thick ZnO plate.
The first, second and third thickness shear modes (TS1...3 )are in-
dicated, together with the fundamental thickness extensional mode
(TE1 ). See Fig. 27 for the nature of the waves.

ent lengths and directions in layers of different materials, they all share the same
x-component and thus lateral wavelength.
The dispersion diagram of Fig. 28 shows the possible combinations of the
lateral component of the wave number βx against frequency as dark lines. It is a
typical case that the dispersion curves form smooth continuous lines. The positive
values of βx stand for real valued wave number, but the negative side of the axis
is used to represent imaginary values of βx .
The lateral wave number βx is real for a laterally propagating loss-less mode.
For the plate modes, the out-of-plane component of the displacement at a given
level in the layer stack, for example on the surface, is given by an expression of
the form
uz = Az ei(βx x−ωt) , (61)
where Az is a complex amplitude. For a real valued βx this is a plane wave prop-
agating in the direction of the x-axis. If, however, βx is imaginary, say
βx = iB, (62)
where B is real, we find
uz = Az e−Bx e−iωt (63)

49
which is an evanescent wave. Thus the evanescent branches of the dispersion
curves represent non-propagating, non-energy carrying vibration. The evanescent
waves have an important role in designing proper boundary conditions in actual
devices.
In a loss-less layered structure surrounded by vacuum, the plate modes have
either purely real or purely imaginary βx . However, if viscous losses are present or
a wave mode radiates energy into bulk waves of an infinite half-space at either end
of the layer stack, the lateral wave number of the plate modes becomes complex.
These modes are no more true loss-less vibration modes but propagating modes
the amplitude of which decays unless energy enters the system in a form of an
exciting force. Handling of complex wave numbers in dispersion calculation is
desirable for assessing the bulk leakage that is indeed present in real SMR type
BAW devices, but adds a significant amount of complexity in the calculation.
Complex wave number simulations have been employed for example in [43]. In
this work, we have limited to pure real or imaginary loss-less wave modes. The
approximation is appropriate for practical cases of SMRs with a high-performance
reflector.

5.2 Lateral Energy Trapping

In a BAW resonator, the vibration needs to be contained in the resonator volume
not only in the vertical direction, where it is done by the Bragg reflector or an
air interface, but also in the lateral direction. This can be accomplished by de-
signing the resonator so that at the resonance frequency, and in a sufficiently wide
band around it, the surroundings support only an evanescent wave. The method is
familiar already from the quartz resonator technology [44].
An important matter to note here is that there are two main types of dispersion
that the thickness extensional resonator can exhibit. The first, often called Type I,
is exhibited by resonators based on ZnO. There the dispersion curve f = f (βx )
of the TE1 mode is monotonically increasing from the cutoff frequency fcutoff ,
defined by the frequency at which the dispersion curve crosses to the imaginary
side (Fig. 29). This means, quite intuitively, that the shorter the lateral wavelength,
the higher the frequency. In such material, the energy trapping can simply be
provided by thinning the outside of the resonator to bring the fcutoff -frequency
of the surroundings higher than that of the resonator, leading to evanescent wave
behavior outside the resonator. The thinning can very simply be done by having a
patterned electrode to define the resonator (Fig. 31, left).
It turns out, however, that the other case called Type II dispersion is much
more common. Especially the most important material of the thin film BAW
technology, AlN, exhibits Type II dispersion. In this case, the TE1 dispersion
curve has a negative slope (Fig. 30) meaning slightly counter-intuitively that the

50
 8
x 10
11

No Electrode
10.5 Electrode
Frequency (Hz)

10

9.5 Trapping Range

Evanescent Propagating (real β )

(imaginary β ) x
8.5 x
−2 −1 0 1 2 3 4 5
Lateral wave number βx (1/m) 5
x 10

Figure 29: Energy trapping illustrated for a Type I resonator provided

by patterned metal electrode. The horizontal lines show the cutoff
frequencies for the electroded and non-electroded areas.

frequency increases when wavelength decreases. In order to bring evanescent

wave behavior to the outside region for such a resonator actually would means
thickening or mass loading the outside area (Fig. 31, right).
Here we have assumed that it is the very same mode inside and outside the
resonator that will with an imaginary wave number provide the energy trapping
boundary condition. This is the case when the thickness/cutoff offset between
the resonator and the outside region is small. If a large step-like discontinuity
is created at the resonator edge, it may bring a completely different plate mode
of the external region to the resonance frequency range, which may also be used
to provide the energy trapping. When the discontinuity is small, the modes in
and out are coupled well together since their vertical shape is almost identical.
However, in the case of a large step and modes of different nature, the situation is
more complex.
The dispersion type for a single plate of certain material is defined by its Pois-
son ratio, or in other terms, by the ratio of the shear bulk wave velocity to the
longitudinal bulk wave velocity r = vS /vL . If this ratio less than 0.5, the mate-
rial exhibits Type I dispersion (rZnO = 0.43), and if it is higher than 0.5, Type II
dispersion (rAlN = 0.56). A ratio of exactly 0.5 leads to a special case of equal
cutoff frequency for the TE1 and TS2 modes.
In Fig. 29 dispersion curves for a simple Type-I structure of Fig. 31 are shown.
Energy trapping is provided by the evanescent wave in the non-electroded region

51
 9
x 10
2

1.95 Electrode
Outside
1.9
Frequency (Hz)

1.85
Trapping Range
1.8

1.75

1.7

1.65
Evanescent Propagating (real β )
(imaginary βx) x
1.6
−4 −2 0 2 4 6 8
Lateral wave number βx (1/m) 5
x 10

Figure 30: Energy trapping illustrated for a Type II resonator provided

by a mass loaded outside region. The horizontal lines show the cutoff
frequencies for the electroded and non-electroded areas.

Figure 31: Left: Two region resonator structure of dispersion Type I

where the patterning of the electrode provides energy trapping. Right:
Two region structure of Type II dispersion where a thickened sur-
rounding is needed for energy trapping.

between the cutoff of the electrode region at about 930 MHz and the cutoff of the
non-electroded region at about 970 MHz indicated as ’trapping range’. Below that
range, there is no propagating wave either in the electrode or outside it. Above
it, the vibration energy is not trapped within the electrode and can radiate to the
surroundings – the device does not function as a high-Q resonator but rather as a
transducer. In the trapping range, there is in principle a perfect reflection of acous-
tic energy at the electrode edges, as the evanescent wave outside does not carry
energy out of the resonator. Thus, high-Q standing wave resonance modes can

52
be formed within the electrode with such lateral wavenumbers that can fulfil the
boundary conditions. Considering a hypothetical case of having a fixed bound-
ary condition uz = 0 at the resonator edges, the standing wave resonances would
occur on the dispersion curve with lateral wave numbers
nπ
βx = , n = 1, 2, 3, ... (64)
W
with W being the resonator width. In practical devices such a simple boundary
condition cannot be realized, and the dispersion characteristics on both sides of
the resonator edge must be considered in order to find the lateral resonances. A
transfer matrix model developed for this purpose by M. Ylilammi at VTT for
a single resonator [34] was adopted for the simulations of Paper 4. The model
was further developed to account for the lateral acoustical coupling of adjacent
resonators and for calculating harmonic response of such laterally acoustically
coupled multiport devices in Papers 5 and 6. An overview of the model is given in
the next section. A detailed description of the multi-port model is given in Paper
5.
Only a single plate wave mode inside and outside the resonator has been con-
sidered here. In reality, there are always other modes with real wave number of
higher magnitude present at any frequency, as is evident from the dispersion di-
agrams of Figs. 37 and 53. These modes will constitute a loss mechanism and
source of additional spurious ripple.

5.3 2-D Modeling of BAW Resonators

It is assumed that the vibration modes of a resonator structure consisting of several
lateral regions i with known dispersion relations can be presented by an out-of-
plane displacement of the form

ui (xi ) = Ai e−jβi xi + Bi ejβi xi , (65)

where Ai and Bi are the amplitudes of right and left propagating waves, respec-
tively, βi is related to frequency by the known dispersion relations βi = βi (f ). By
requiring continuity of the displacement ui and its derivative ∂u i
∂xi
(stress) over the
region boundaries, one can form a transfer matrix
" # " #
Ai+1 Ai
= Mi (66)
Bi+1 Bi

to relate the displacement amplitudes in adjacent regions. This allows for a nu-
merical solution of the laterally standing wave resonances up (x) of the structure.
The set of eigenmodes up (x), p indexing the modes by eigenfrequency ωp , is used

53
 Figure 32: Illustration of an exemplary 2-D model geometry for a
five region acoustic device having two electrical ports. The gray area
at the bottom reflects the topology: the elevated regions (i = 2, 4)
have electrodes and their dispersion characteristics are described by
the curve labeled ’Electrode’ in Fig. 29, whereas the other regions
(i = 1, 3, 5) are described by the curve labeled ’No Electrode’. The
electroded region 2 is defined to form the electrical port 1 and region
4 port 2. The boundary conditions for the amplitudes of the energy
trapped modes are shown.

as a basis function set to describe the harmonic response of the system. In Fig.
32 an exemplary five-region resonator illustrating the model setup is shown. It is
to be noted that within this model, the eigenmodes depend only on the dispersion
characteristics and geometry of the regions, while the electrical drive conditions
can be freely set by simply choosing regions that represent the electrodes. Several
electrical ports can be considered simultaneously.
The electromechanical coupling coefficient kp of a mode up can be calculated
with respect to a driving voltage present at regions i that belong to an electrical
port m from ¯ ¯
¯P R ¯2
¯
¯ i∈port m up (x) dx¯¯
¯ Wi ¯
2 2
kp,m = k1D +∞
, (67)
R
Wx,port m |up (x)|2 dx
−∞

where k1D is the effective coupling coefficient of the resonator obtained from 1-
D simulation, Wx,port m is the total width of the electrode(s) belonging to port m.
Equation (67) follows from (35) with the assumption of constant electric field over
the electrode of port m.
Having solved the eigenmodes up and their coupling coefficients, assuming
realistic Q-values for the modes (Q-values do not emerge from this model), one
can calculate the harmonic response of the system as an eigenmode superposition.

54
Furthermore, this can be done considering any defined region or set of regions act-
ing as the driving electrical port allowing for simulation of the characteristics of
an N -port device with acoustic coupling between the ports. The detailed deriva-
tion of the harmonic response is presented in Paper 5 while the main results are
summarized below.
The harmonic vibration response of the device to a driving voltage V present
at port m of the device is
X
ux,m (x, ω) = αp,m Fp (ω, ωp )up (x), (68)
p

where Fp (ω, ωp ) is the shape of the modal frequency response

1
Fp (ω, ωp ) = Qp (ωp2 −ω 2 )
, (69)
1+ jωωp

from (18) and αp,m is the modal superposition coefficient

2
V Kp,m Qp Wx,port m εS
αp,m = R . (70)
je Wport m up (x) dx
2
Here Qp is the Q-value associated to mode up , V is the driving voltage, Kp,m =
2 2
kp,m /(1 − kp,m ) is the coupling of mode p to port m, Wx,port m is the total width of
the electrodes of port m, and e and εS are the piezoelectric stress coefficient and
the permittivity of the piezolayer, respectively. Integration is over all electroded
regions belonging to port m.
On the other hand, if one knows the vibration response ux,m (x, ω) resulting
from a harmonic driving voltage V at port m, the current trough port n is obtained
from Z
jωeWy,n
Inm = jωV C0,n δnm + ux,m (x, ω) dx. (71)
t
Wx,n

Here C0,n is the plate capacitance formed by port n, Wy is the width of the elec-
trode in the truncated dimension and t the thickness of the piezolayer. The driving
voltage V is assumed to be zero at all other ports than the driving one (hence the
Kronecker delta δmn ), i.e., they are short circuited. This allows for the extraction
of the N -port Y -parameters of the device
Inm
Ynm = . (72)
Vm
This constitutes a full electrical description of the acoustic device with multiple
acoustically coupled electrical ports

55
 929.0377 MHz 929.3301 MHz 929.8334 MHz
k=0.209 k=0 k=0.069
1 1 1

0.5 0.5 0.5

u(x) (a.u.)

u(x) (a.u.)

u(x) (a.u.)
0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 0 1 2 3 0 1 2 3
x (m) x 10
−4 x (m) x 10
−4 x (m) x 10
−4

930.529 MHz 931.4133 MHz 932.4802 MHz

k=0 k=0.04 k=0
1 1 1

0.5 0.5 0.5

u(x) (a.u.)

u(x) (a.u.)

u(x) (a.u.)
0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 0 1 2 3 0 1 2 3
x (m) x 10
−4 x (m) x 10
−4 x (m) x 10
−4

933.7225 MHz 935.1322 MHz 936.7008 MHz

k=0.027 k=0 k=0.02
1 1 1

0.5 0.5 0.5

u(x) (a.u.)

u(x) (a.u.)

u(x) (a.u.)

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 0 1 2 3 0 1 2 3
x (m) x 10
−4 x (m) x 10
−4 x (m) x 10
−4

Figure 33: The first 9 eigenmodes of a simple ZnO resonator where

only the electrode is patterned to be 308 µm wide. The large width of
the resonator leads to closely spaced resonance modes. The coupling
to the antisymmetric modes is zero.

As an example, the first 9 eigenmodes of a 932 MHz one-port resonator such

as the one presented in Fig. 31 are shown Fig. 33. In total, such a wide res-
onator supports a total of 22 eigenmodes in its trapping range, of which 11 are
symmetric and have non-zero coupling. The energy trapping is evident from the
exponentially decaying vibration amplitude at the outer regions. In Fig. 34, the
eigenfrequencies of an otherwise identical but only 150 µm wide resonator are
indicated on the dispersion diagram and on the input impedance obtained from
the harmonic response calculation. It can be seen that the eigenfrequencies corre-
sponding to symmetric modes coincide with local minima in the impedance while
the antisymmetric ones do not affect it. The smaller width of the resonator has
been chosen in order to make the spurious ripple spectrum more sparse for illus-
trative purposes.
The described 2-D model gives a strongly simplified picture of the resonator. It

56
 8 8
x 10 x 10
9.8 9.8
Impedance Dispersion data

9.7 9.7 No Electrode

Electrode

9.6 9.6
f (Hz)

9.5 9.5 Trapping range

9.4 9.4

9.3 9.3

9.2 −4 −2 0
9.2
10 10 10 −2 −1 0 1 2 3 4
|Z| (Ω) βx (1/m) 5
x 10

Figure 34: Calculated eigenmode frequencies (horizontal lines), dis-

persion diagram (right), and magnitude of input impedance (left) of a
simple ZnO resonator where only the electrode is patterned to be 150
µm wide. A total of 11 modes are supported in the trapping range be-
tween the dispersion curves. The eigenfrequencies of the symmetric
modes are associated with local minima in the impedance.

considers only the vertical displacement of a single plate mode, and is based on the
requirement of continuity of the displacement and its first derivative. This limits
its applicability to structures with only small discontinuities between the lateral
regions. It omits the fact that proper continuity conditions should be dealt with
along the vertical interface between regions through the entire layer stack, as is
done by Milsom et al. in [43]. It is expected that this model is not realistic enough
for example in the case of heavy electrodes causing large acoustic discontinuity at
the edges. The advantage of the this 2-D model is that it is very fast and intuitive
and it has shown its usefulness in rough analysis of the principles of the device
operation. The 2-D model can also be extended to two lateral dimensions with the
aid of numerical eigenmode solution [45].

5.4 FEM Modeling of Thin Film BAW Devices

Fully numerical solution of electro-acoustic equations by for example the Finite
Element Method (FEM) allows one to abandon many of the simplifying assump-
tions used in the 1-D and 2-D models discussed in Sections 3.1 and 5.3.
In Papers 3 and 4, the finite element software developed specifically for the

57
thin film BAW acoustic devices by T. Makkonen et al. (Materials Physics Labo-
ratory, Helsinki University of Technology) called BAWFEM [46] was used.
Advantages of FEM over the simpler models are numerous. First of all, more
realistic resonator geometries can be modeled. A FEM model also intrinsically
contains the dispersion characteristics for all plate modes that the applied element
mesh can represent. The vertical boundaries between regions of the resonator hav-
ing different layer stacks are handled naturally without simplifying assumptions,
even in the case of large discontinuities. The electrical excitation can be defined
realistically and the effects of the fringing fields are included. Coupling to differ-
ent plate modes and shear waves arises naturally. Also, full anisotropy and viscous
losses of the materials can be handled. Ohmic losses in the electrodes can also be
modeled by FEM [47], but this has not been implemented in BAWFEM nor em-
ployed in this work. As will be discussed later, it is actually in the modeling of
losses where the greatest difficulties and drawbacks of FEM are encountered.
The computation with FEM becomes easily quite heavy. In the typical models
for 1–2 GHz frequency range thin film BAW resonators, the element size needs to
be around 1 µm in the lateral direction and even smaller in the vertical direction in
order to be able to represent effects related to the high-lateral-wave-number plate
modes. While this is fairly easily managed for 2-D models, realistic 3-D modeling
becomes extremely heavy.
In Fig. 35 the geometry of a 2-D FEM model of the vertical cross-section of a
ZnO resonator employed in Papers 4 and 3 is illustrated. The degrees of freedom
to be solved are the displacement components (ux , uz ) and the electric potential
φ, for which the boundary conditions have also to be specified.
Symmetry is utilized, so that only the left half of the cross section of the res-
onator is included in the model. The symmetry condition is applied by setting
the lateral displacement to ux = 0 at the symmetry edge. At the top surface, no
conditions are set for the displacements, i.e., the surface is free. At the bottom
the free condition is also used, but this is realistic for a device sitting on a sub-
strate only when the model is truncated at the lower edge with a high-Z material.
This creates a stress node at the interface, which is a good approximation to a
free surface. However, no energy at all can leak out of the bottom in the finite
model, so the finite reflectivity of the reflector cannot be analyzed by this kind of
boundary conditions. At the left edge outside the resonator, the layer structure is
truncated at a distance large enough for the exponentially damped tails of wave
modes trapped by the electrode to diminish. There both ux and uz are clamped to
zero. For the trapped wave modes the boundary condition makes no difference,
but for the non-trapped high-βx modes visible in Figs. 37 and 53, it does have a
large effect. This is a source of error in the simulations.
The electrical potential φ has to be specified in all of the conductors of the
model. The excitation is specified via having a harmonic voltage of 1V in the

58
 Figure 35: An example of a 2-D FEM model of an ZnO SMR. The
model consists of the left half of the resonator, utilizing symmetry
condition at the right edge. The boundary condition assumes that
there is free space on the top of the resonator as well as at the bot-
tom. Truncating the model at a high-Z layer makes the latter a good
approximation to the real situation. At the left edge, i.e. outside the
resonator, a fixed boundary condition is used. For wave modes trapped
in the electroded region, this boundary condition makes no difference,
but for non-trapped waves it is of significance. Excitation is provided
by a voltage difference between the electrodes.

upper electrode and the lower electrode grounded. The metals in the reflector
are in this case defined grounded but they could as well be left floating with the
condition that the potential is constant over any connected metal region.
The material parameters are specified via the stiffness matrix cE ij , density, di-
electric permittivity εS , and piezoelectric stress matrix eij , and a viscosity matrix
ηij . If the full anisotropic matrices are not known, an isotropic approximation is

59
 0
10

−1
10
k eff, Y ( Ω 1)
−
10 µm

−2
10

−3
10

−4
10
9.2 9.3 9.4 9.5 9.6 9.7
f (Hz) x 10
8

Figure 36: FEM-simulated eigenmode spectrum (keff,p vs. fp ) and

corresponding resonator admittance Y (f ) for ZnO SMR model such
as the one presented in Fig. 35. The FEM calculation resolves the
same modes as the 2-D model but also a large number of additional
modes.

easy to construct based on acoustic wave velocities vL , vS and density ρ.

There are two methods of solution for the specified FEM problem: In the
modal analysis the loss-less eigenmodes of the system are solved and in harmonic
analysis the response of the system to harmonic driving (typically voltage applied
over the electrodes) at any specified frequency is calculated.
The modal analysis results in free eigenmodes with no attenuation. Each mode
is associated with a solution for the field variables ux (x, y), uz (x, y), φ(x, y), and
the eigenfrequency ωp together with a modal constant describing the coupling co-
efficient of the mode to the specified exciting electrodes. The modal constants are
calculated by integrating the accumulated charge over the interface of a specified
electrode and the piezoelectric layer for each eigenmode. If one additionally as-
signs each mode a Q-value (which does not result from loss-less modal analysis),
one can via the equivalent circuit analogy extract the harmonic electrical response
of the resonator provided that a large enough number of eigenmodes have been
solved.
This yields for a computationally efficient method for computing the harmonic
response via modal analysis with the drawback that the viscous losses are not
incorporated properly. This, however, is not as large a drawback as it may first
seem.
In Paper 4, the modal analysis was utilized together with the 2-D model to
analyze the spurious resonance suppression in ZnO resonators. An example of
a FEM-calculated eigenmode spectrum and the resulting frequency response are

60
shown in Fig. 36. The 2-D FEM predicts the same strong modes as the 2-D
model, but also a large number of additional modes (See Paper 4 for a detailed
comparison).
The harmonic analysis incorporates losses via the viscosity of the materials.
This improvement over the lossless modal analysis, however, turns out not to be so
significant. The Q-value of the typical devices does not seem to be dominated by
the viscous losses, and the dissipation properties of the thin film materials at GHz
range frequencies are poorly known anyway. The harmonic analysis, however,
allows one to calculate the vibration field ~u(x, y) = [ux (x, y), uz (x, y)] and the
potential field φ(x, y) at any frequency which corresponds to the actual situation
of driving such a resonator under, for example, a vector network analyzer or laser
interferometric measurement.
An interesting post-processed result allowed by the harmonic analysis is the
extraction of the dispersion characteristics by Fourier techniques in the same man-
ner as for the laser interferometric measurements described in Section 6. An ex-
ample of such FEM-simulated dispersion characteristics is presented in Fig. 37.
This was utilized in Paper 3 for comparing simulation and measurement of the
reflector transmission characteristics as function of frequency and wave number.
The most important source of error as compared to real devices comes from
the boundary conditions at the model edges. The FEM model is necessarily fi-
nite in size, and some conditions must be fixed for the acoustic fields at the outer
boundaries. In real devices, some acoustic energy can escape the resonator in
the form of acoustic radiation. The finite FEM model on the other hand cannot
lose any energy except in viscous dissipation. The acoustic leakage, which is in
practice a more important factor for defining the Q-value than the viscous losses,
is then not modeled at all leading to unrealistically high Q-values predicted by
FEM modeling (unless viscous loss dominates). This problem has been tried to
be solved by creating a padding of viscous material around the resonator in an
attempt to absorb most acoustic radiation from the active resonator model [48]. A
profile ramping the viscosity up over a distance must be used in order not to create
reflections from abruptly changing viscosity. In the vertical direction, where the
vibration is close to being purely longitudinal, the method has been rather suc-
cessfully applied. Problems arise when trying to generate a viscous profile in the
lateral direction, that would work simultaneously for all wave modes and lateral
wavelengths propagating along the layers. This method also makes the models
larger and thus computationally heavier.
A special kind of mathematically modified computation element attempting
to suppress the boundary reflections called the perfectly matched layer (PML)
is available in some FEM simulation software. Mayer et al. report in [49] on
the usage of the PML in the case of SAW simulation and find it superior to the
absorbing boundaries. The simulation of SAW devices is slightly simpler than

61
 1200
80
1100
TE1
1000 60
Frequency (MHz)

900
40
800
TS2
20
700

600 0
TS1
500
−20

400
−40
0 0.5 1 1.5 2 2.5 3
(dB)
|β||| (1/µm)

Figure 37: The dispersion diagram calculated from the out-of-plane

surface displacement simulated by 2-D FEM harmonic analysis for a
ZnO SMR resonator. The first thickness shear (TS1 ), second thick-
ness shear (TS2 ) and the first thickness extensional (TE1 ) mode
branches are indicated in the plot.

that of BAW devices since there periodicity can be utilized - the PML needs only
be applied in the vertical direction.
A third kind of approach is to use a combination of FEM and the boundary
element method (BEM), where the whole substrate or chip can be incorporated
as in BEM only the outer boundary of the model is discretized, thus keeping the
computational size of the substrate-containing model manageable [50]. The FEM-
BEM is, however, computationally much more involved than plain FEM [49]. It
also seems that the FEM-BEM is difficult to set up for the lateral edges of the res-
onator. To the author’s best knowledge, no perfect solution exists for the boundary
conditions in the numerical simulation of BAW resonators, but the developments
related to using the PML seem quite promising.
Despite the difficulties in obtaining high level of realism in the FEM calcula-
tion due to the issues discussed above, 2-D FEM has been very useful in the practi-
cal work for optimizing resonator performance – mainly in the analysis of spurious
resonance behavior. Qualitative agreement with experiment is good enough to fa-

62
cilitate the use of FEM in a numerical experiment to verify various designs before
entering sample fabrication. Often the improvements to the designs have come
from simpler ideas and models such the dispersion simulation, or 2-D simulation,
and the final test before experiment is performed by FEM.

5.5 Spurious Resonance Suppression

Spurious ripple due to standing wave resonances of laterally propagating plate
waves is a direct consequence of lateral energy trapping. A filter constructed of
resonators showing such spurious response will have unacceptable ripple in the
passband (see Fig. 41).
Several methods exist for manipulating the spectrum of spurious resonances.
First, the lateral dimensions affect the spurious spectrum so that the smaller the
resonator width, the larger the frequency spacing of the resonances. In principle,
the resonator can be made so small that only the principal mode fits in the trapping
range. However, for practical applications the required impedance level (typ. 50
Ω) sets a strict requirement for C0 and thus the area of the resonators making this
not an option. Making the resonator size very small would also lead to trade-off in
coupling and the Q-value due to a significant fraction of vibration energy lying in
the evanescent fields outside the electrode-covered area and increasing perimeter
to area ratio, respectively.
Secondly, one can reduce the lateral symmetry of the resonator since obvi-
ously the resonances are most pronounced in resonators of high symmetry such
as circular or square resonators. In [51], a method of spurious resonance suppres-
sion based on symmetry reduction, or ’apodization’, is presented. The effect is to
make the spectrum of the modes to split into a large number of weakly coupled
modes instead of a smaller number of strongly coupled modes. Elliptic shape has
also been reported as advantageous for spurious suppression [52]. This has also
been studied in [45] by investigating the spurious resonance spectra of ellipses
with varying aspect ratio.
A third possibility is to engineer the resonator to have a very flat dispersion
profile where all the spurious resonances would be spaced very close to each other
in frequency, practically merging together in the electrical response [53], [54].
The dispersion engineering is quite demanding to realize, as there are many op-
timization goals to be reached simultaneously and not very much freedom in the
choices of materials and their thicknesses.
In [23] another method for spurious resonance suppression based on modifi-
cation of the lateral acoustic boundary condition is presented. This method, the
boundary frame method, is very effective provided that the dispersion characteris-
tics of the resonator can be tailored to make it applicable. It is based on modifying
the spurious resonance modes so that only the principal mode can couple to the

63
 Figure 38: Cross section of a resonator with the boundary frame struc-
ture for spurious resonance suppression. The dashed line on top of the
resonator shows the cutoff frequency at each region of the resonator.

driving field and the others are extremely weakly coupled to it. A detailed study
of ZnO resonators employing this method and comparison to 2-D and 2-D FEM
models was conducted in Paper 4 to analyze the boundary frame method in depth.
In the boundary frame method one creates a thickened frame of a certain width
(see Fig. 38) with a lowered cutoff frequency at the resonator edge so that a
free boundary condition is simulated – the boundary region acts as a matching
structure between the resonator and the surroundings. Calculated dispersion char-
acteristics for such a resonator are presented in Fig. 39. As a result, the fun-
damental mode takes a shape with constant amplitude over the resonator which
closely matches to the shape of the driving electrical field. If the higher order
lateral modes are orthogonal to the fundamental mode, a property indeed often
encountered in eigenvalue problems, the fundamental mode will be the only mode
coupled to the driving field.
The boundary frame dimensions have to be optimal for given dispersion char-
acteristics of the outside area, resonator area, and the frame area itself. In Fig. 40
electrical measurements of varying boundary frame widths for a 930 MHz ZnO
resonator are shown. A drastic reduction on the spurious ripple is observed for
a certain width. Measurements of pass band transmission in filters consisting of
similar resonators having a varied frame width are shown in Fig. 41. The ad-
vantage of using an optimal boundary frame is evident in the plots labeled 9 µm
and 11 µm, indicating the frame width. The resonators of these filters are slightly
different in design than those in Fig. 40, thus the difference in frame widths.
The more detailed structure of the resonator of this study was presented in
Fig. 6 (left), where it can be seen that the boundary frame is formed by letting the
bottom electrode and an insulator layer to overlap slightly.

64
 1100
1080
1060 boundary frame
1040
center area
1020
outside
f (MHz)

1000
980 fc,outside
960
940 f
c,center

920 f
c,boundary

900
880
−2 −1 0 1 2 3 4 5
β (1/m) 5
x 10
x

Figure 39: Calculated dispersions for a 930 MHz ZnO resonator uti-
lizing the boundary frame method. The energy trapping range is indi-
cated by a vertical arrow next to the βx = 0 axis. The boundary frame
supports a propagating wave throughout the trapping range and acts
as a matching element between the outside and the resonator center
area.

To confirm that the mechanism behind the suppression of spurious resonances

is indeed the one described above, the mode spectrum obtained by the 2-D model
and 2-D FEM was studied in Paper 4. It was observed that when a piston mode
type mode emerges as one of the lateral modes, the coupling of the others is indeed
reduced on a wide frequency range. However, the coupling coefficients of the
spurious modes do not reduce to zero neither in the 2-D nor the FEM model. A
calculation of the inner products among the eigenmodes of the 2-D simulation
shows that the modes have a very small level of linear interdependence but no
perfect orthogonality. A likely reason for the lack of orthogonality is that the
boundary condition for the eigenmodes is not constant over the frequency range
due to dispersion but is slightly different for each mode. Another effects was
found to affect the spurious mode content. Depending on the frame width, a
single mode was strongly suppressed while others were not. It was found that this
’missing mode’ effect is due to the frame region and the center region forming for
a certain frequency a symmetric modeshape for which the coupling becomes close
to zero similarly to the antisymmetric modes. The minimum amount of ripple
results from the interplay of the two effects and does not necessarily coincide
exactly with the piston mode operation point but is typically very close to that.

65
 3.5 µm 5.5 µm
1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1

7.5 µm 9.5 µm
1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1

Figure 40: Measured scattering parameter S11 on the Smith chart for
ZnO resonators with varied boundary frame width. The frame width
is indicated on top of each chart. The resonator with 5.5 µm frame
shows almost perfectly smooth response except for at near the bottom
of the chart.

In Fig. 42, the harmonic vibration response of a resonator simulated with the
2-D model is presented as a function of frequency for a resonator with no bound-
ary frame (top) and for a resonator with an optimal boundary frame (bottom). The
flat amplitude at resonance (fs = 930 MHz) and clearly smaller contribution of
spurious modes above fs is evident in the latter.
The discussion has concentrated on resonators exhibiting so called Type I dis-
persion only. Type I dispersion refers to ZnO-type behavior where the TE1 dis-
persion curve has a positive slope, i.e., the lateral wavenumber increases with
frequency. Most materials, especially AlN, show intrinsically just the opposite
behavior – the TE1 dispersion curve has a negative slope when starting from the
βx = 0 axis for a certain region, before turning upwards. In this case the bound-

66
 0

−5 7 µm
9 µm
IL (dB)

11 µm
13 µm

−10

−15
940 950 960 970 980 990
f (MHz)

Figure 41: Three-stage ladder filters comprising of resonators with

different frame widths. Both 9 and 11 µm-cases show very smooth
passband, whereas on the 7 and 13 µm curves significant ripple is
observed.

ary frame structure must be inverted, i.e., a recessed trench must be created in the
place of the frame (Fig. 43). To avoid practical difficulties associated with this,
a method of dispersion flipping in the SMR devices was developed by Infineon
Technologies GmbH for making AlN-resonators to behave like ZnO-resonators
and to apply the boundary frame [55]. The boundary frame method is nowadays
being used in some form by most of the major industrial manufactures of BAW
resonators.

67
Figure 42: Logaritmic plot of the vertical displacement vs. frequency
simulated by the 2-D model for a resonator without a boundary frame
(top) and a resonator with a piston-mode producing boundary frame
(bottom). Each horizontal line of the plot represents the vertical dis-
placement amplitude vs. lateral position along the surface of the res-
onator. Thin vertical lines show the electrode edges (top) and the
boundary frame (bottom).

68
Figure 43: Cross section of an exemplary realization of the boundary
frame method for a Type II resonator. The dashed line on top of the
resonator shows the cutoff frequency at each region of the resonator.

69
5.6 Laterally Acoustically Coupled Resonator Filters
In Fig. 44 the fundamental and the only trapped resonance mode of a rather small
resonator is shown. The vibration is energy-trapped by the electrode-induced cut-
off frequency difference and the evanescent tail extending to the surrounding re-
gion is clearly visible. Consider bringing two such resonators in such a close prox-
imity to each other, that their evanescent tails extend to the other resonators region
with significant amplitude. The two resonators will form a coupled-resonator sys-
tem supporting a symmetric and an antisymmetric mode (Fig. 45). The device
acts as a pass-band filter with acoustical signal coupling, the width of the pass-
band being defined by the frequencies of the two modes. It is exactly this kind of
devices that motivated the development of the multiport 2-D model development
discussed in Section 5.3 and in Papers 5 and 6.

W=7 µm
1
0.5
u(x) (a.u.)

0
−0.5
−1
−15 −10 −5 0 5 10 15 20
x (µm)

1940

1900
f (GHz)

1860

1820
−0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
β (1/µm)
x

Figure 44: Simulated displacement (top) and frequency of the sole

eigenmode and dispersion characteristics (bottom) of an isolated nar-
row resonator. In the displacement plot the vertical lines indicate the
limits of the electroded area. In the dispersion plot the resonance fre-
quency is indicated by a horizontal black line. The upper gray curve
is the dispersion curve of the non-electroded outside region and the
lower curve is for the electroded central region.

The acoustical coupling of adjacent resonators, referred to as the monolithic

crystal filters (MCF, Fig. 46 A) has been utilized in quartz-crystal based technol-
ogy already since 1960’s to form narrow-band filters up to frequencies of some
tens of MHz with galvanically isolated input and output [56, 8]. Another possibil-

70
 W=7 µm G=5 µm
1

u(x) (a.u.) 0.5

0
−0.5
−1
−15 −10 −5 0 5
x (µm)

1940

1900
f (GHz)

1860
S: 1878 MHz
1820 A: 1887 MHz
−0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
βx (1/µm)

Figure 45: Simulated displacement (top) and frequency of the two

eigenmodes and dispersion characteristics (bottom) of system of two
narrow electrode regions in close proximity so that the resonators are
coupled acoustically via the evanescent wave in the gap between them.
In the displacement plot the vertical lines indicate the limits of the
electroded areas. In the dispersion plot the resonance frequencies are
indicated by horizontal black lines. The upper gray curve is the dis-
persion curve of the non-electroded outside and gap region and the
lower curve is for the electroded resonator regions.

ity to realize an acoustically coupled filter is to stack two resonators on top of each
other to form a stacked crystal filter (SCF, Fig. 46 B), and a further development
of this, the coupled resonator filter (CRF, Fig. 46 C) having a layer or layers in
between the two resonators to control the amount of inter-coupling for the res-
onators and providing means for galvanic isolation. The SCF and the CRF have
been realized utilizing AlN thin film based BAW technology by several groups
[57, 58, 59, 60, 61] motivated by the wide bandwidth, very small footprint, and
possibility for balanced-unbalanced conversion. The fabrication of the vertically
stacked thin film BAW resonators is, however, difficult to do economically due to
process complexity and extreme tolerance requirements of fabrication.
The laterally coupled thin film BAW resonator filters [62, 63, 64, 65] (Fig. 46
D), a direct analogy of the MCF, are an attempt to obtain the benefits of the SCF
and CRF with a simplified manufacturing technology and less stringent fabrication
tolerances.
In its simplest form, the LBAW filter is formed by just two electrode areas

71
 Figure 46: Illustrations of some acoustically coupled resonator fil-
ters: A) the monolithic crystal filter (MCF) B) the stacked crystal fil-
ter (SCF) C) the coupled resonator filter (CRF) and D) the laterally
acoustically coupled filter (LBAW).

with a narrow gap between them. A study of this kind of a filter prototype was
conducted in Paper 5. The bandwidth of the filter can be affected by the geometry
(widths of the electrodes and the gap between them). To maximize the bandwidth,
one must try to make the symmetric mode to have as low frequency as possible
and the antisymmetric mode as high frequency as possible (we limit the discussion
to Type I dispersion here). This is achieved by making the electrodes and the gap
as narrow as possible while still supporting the two modes. The hard-wall limit
for the frequency difference is the width of the energy trapping range. Insertion
losses of two-electrode filters with varied geometry are presented in Fig. 47.
By a very simple two-resonator design fabricated by a single lithography pro-
cess, pass band filters can be demonstrated, but with a modest pass band width
(up to 1.45 %) and parasitic effects such a spurious pass band below the intended
one (Papers 5 and 6). A photograph of such experimental device is shown in Fig.
48. In order to create filters with pass bands wide enough to be useful for e.g. the
modern mobile communications devices, one can multiply the number of elec-
trodes creating an interdigital type of structure similar to that in SAW devices.
The effect of this is to press down the frequency of the symmetric mode as it is
dependant on the total width of the device. The frequency of the antisymmetric
mode is meanwhile increased as compared to that of the two-electrode basic ele-

72
 Insertion loss (dB)
0

10

20
IL (dB)

30

40 W=4µm G=1µm
W=4µm G=5µm
W=7µm G=1µm
50
W=7µm G=5µm

60
1800 1850 1900 1950 2000
f (MHz)

Figure 47: Insertion losses calculated for a two-electrode LBAW filter

with various widths of the electrodes and gap. The combination of
narrow fingers (W = 4µm) and narrow gap (G = 1µm) yield the
widest bandwidth.

ment as the wavelength is determined to a higher extent by the periodicity of the

electrode structure and less by the evanescent fields outside the filter. In between
the two main modes, intermediate modes are also supported. A photograph of a
31 finger device is presented in Fig. 49 and the insertion loss of the device in
Fig. 50. Post-measurement matching by with 120 Ω system impedance and 5 nH
parallel inductances at both ports has been applied.
The device designed by J. Meltaus at VTT with the simulation tool developed
in Paper 5 is remarkable for the following reasons:

• It shows a pass band with a relative width of 4.9 % which is competitive

with commercial SAW and BAW microacoustic filters.

• It has steep passband edges as required by the mobile communications ap-

plications.

• It utilizes standard SMR BAW fabrication technology for the layer stack
and just a single lithography with dimensions much more relaxed than in
SAW at similar frequency range.

• Its footprint is merely that of a single BAW resonator used in ladder filters.

73
 Figure 48: Photograph of a two finger LBAW filter.

Figure 49: Photograph of a 31 finger LBAW filter.

Challenges remaining for the LBAW technology include removing the ripple
in the passband, suppressing the parasitic passband, improving the stopband at-
tenuation and designing the filters so that they can be matched to 50 Ω. These im-
provements will most likely require more mask layers to be added in the process,
but on the other hand it seems feasible that the LBAW technology could allow for
balun functionality and bandwidths even wider than 5 %. These demonstrated,
the LBAW has potential to become a disruptive technology for RF filtering.

74
 0

−10

−20
S21 (dB)

−30

−40

−50

−60
1700 1800 1900 2000 2100 2200 2300
Frequency (MHz)

Figure 50: Transmission S21 of a 31 finger LBAW filter based on AlN

SMR technology and just a single lithography step. Post-measurement
matching with 120 Ω system impedance and 5 nH parallel inductances
at both ports has been applied.

75
6 Interferometric Imaging of BAW Resonators
Scanning laser interferometry provides a method to image surface vibrations of
microacoustic devices with high sensitivity and dynamic range [66, 67, 68, 69,
70]. The method is based on splitting a laser beam into two paths, of which the
first one is reflected from the vibrating surface of the microacoustic device and the
other one serves as the reference beam. The interference of the recombined beams
is used to obtain information on the surface vibration. The sample is scanned rel-
ative to the fixed laser beam that is focused to a small spot roughly 1µm in size,
yielding for a 2-D map of the vibration field. The technique has been intensively
developed at the Helsinki University of Technology since the 1990’s. Two dif-
ferent setups have been employed in the papers of this work: a Michelson-type
interferometer [71] capable of measuring the amplitude of vibration and a later
development, a heterodyne interferometer [72] yielding for both absolute ampli-
tude and phase data of the out-of plane vibration. As small as 1 pm amplitudes
can be detected, the applicable frequency ranging up to 6 GHz. The heterodyne
setup is presented in Fig. 51.

Singlemode
HeNe -laser

Optical
Isolator
Collimation
Optics
PD

AOM Polarizer Microscope

M objective
I
M 0
PBS λ/4
Sample

Figure 51: Schematic of the scanning heterodyne interferometer used

for imaging acoustic vibrations of GHz frequency devices. The sam-
ple is scanned in front of the measurement beam in order to form a
spatial map of the vibration fields [72].

76
 pm f = 932 MHz f = 934 MHz f = 936 MHz
280

106

40

15
6
pm f = 938 MHz f = 944 MHz f = 948 MHz
280

106

40

15
6

Figure 52: Interferometrically measured out-of-plane vibration am-

plitudes of a 932 MHz ZnO SMR at the series resonance frequency
and at selected frequencies above it. The laterally standing wave reso-
nances are pronounced at the higher frequency scans. Image courtesy
of Kimmo Kokkonen, Aalto University.

The interferometer can be used to image direct spatial wave fields in order to
study the vibration modes of the BAW resonators. It can be used to verify that
the devices are operating as designed and modeled such as in Paper 5, and to
analyze spurious modes. In Fig. 52, measured vibration fields of a 932 MHz ZnO
BAW resonator at the series resonance frequency and at some selected frequencies
above it, coinciding with spurious oscillations, are presented.
By measuring the vibration fields on the surface of the device as a func-
tion of frequency, one can extract the lateral dispersion characteristics by using
Fourier techniques [73]. This includes first windowing the measured 2-D dis-
placement field u(x, y; f ), applying 2-D discrete Fourier transformation to yield
a spatial spectrum U (βx , βy ; f ), where βx and βy are the components of the lat-
eral wavenumber, and then reducing the data to a single wavenumber q dimension
U (β|| , f ) for example by averaging along circles of radius β|| = βx2 + βy2 . An
example of a high resolution dispersion diagram measured by laser interferometry
is shown in Fig. 53.
In [74] the correspondence of the ripple in the electrical response and inter-
ferometrically observed mechanical response, integrated over β|| to yield a power
spectrum was presented, proving experimentally the connection of the laterally
standing wave modes and the spurious ripple in the electrical response (Fig. 54).
In Paper 1, the measured dispersion curves were used to find information on

77
 1200 60

1100 50
TE1
40
1000
30
Frequency (MHz)

900
20
800
TS2 10

700 0

600 −10
TS1
500 −20

−30
400
−40
0 0.5 1 1.5 2 2.5 3 (dB)
|β||| (1/µm)

Figure 53: Dispersion diagram measured by a laser interferometer

from a ZnO SMR device. Image courtesy of Kimmo Kokkonen, Aalto
University.

the material parameters of the thin films used in the device by fitting a dispersion
model to the measurement. This allows in principle the extraction of full material
parameters for both longitudinal and shear wave modes, provided that the film
thicknesses are known accurately. While the initial attempt was in the quantitative
sense not so successful, most likely due to inaccuracy in the layer thicknesses, the
method was later quite successfully adopted by A. Link et al. in [75].
In optimizing the performance of BAW resonators and devices, the accurate
knowledge and control of dispersion characteristics is a key issue. While the dis-
persion can in principle be modeled rather easily, it requires accurate knowledge
of material parameters and layer thicknesses. Interferometric measurement pro-
vides a way to directly measure whether the design and modeling has succeeded
and give feedback to the selection of the model parameters. Dispersion measure-
ment can also be used as diagnosis tool for ill-working resonators.

78
 0

-2
Absorbed power (1-|S11|2) (dB)

Mechanical response (a.u.)

-4

-6

-8

-10

-12

920 925 930 935 940 945 950 955 960 965 970
Frequency (MHz)

Figure 54: Comparison of absorbed electrical power 1 − |S11 |2 (solid

line) and mechanical response integrated over spatial wave number
(dashed line) both showing the spurious oscillations with high level of
correlation [74].

79
7 Loss Mechanisms in a BAW Resonator
The loss mechanisms in the thin film BAW resonators include leakage to the sub-
strate, laterally escaping waves, ohmic losses, viscous losses, dielectric losses, and
wave scattering. Also mentioned in the literature are eddy current losses [6] and
losses due to formation of conducting channels in the high resistivity Si substrate
– SiO2 interface [76].
Acoustic leakage to the substrate is characteristic to the SMR type devices
and is almost non-existent in FBARs suspended on air. This is due to the acous-
tic reflector having only finite reflectivity as compared to the practically perfect
substrate isolation of FBAR type devices.
Laterally escaping waves are observed in the interferometric measurements
and FEM simulations [6, 75]. They arise from excitation of higher βx plate wave
modes that are not trapped within the resonator. The reason for their excitation is
most likely the discontinuity at the resonator edge.
The ohmic losses include the resistance of the electrodes but also ohmic dis-
sipation of locally induced charge concentrations associated with the vibration
patterns of the spurious modes. Utilizing low resistivity materials such as Au or
Al in the electrodes with high enough thickness reduces ohmic losses, but resis-
tivity must be co-optimized with other properties such as the coupling coefficient
and this easily leads to trade-offs [77, 32].
Viscous losses are intrinsic to the materials used in the devices. Apart from the
choices of materials, not much can be done about them. Among the acoustic ma-
terials typically used in BAW devices, SiO2 and Al have largest losses, while AlN,
Mo, and W are all rather low-loss materials. Viscous dissipation is proportional
to the space derivative of strain, so one may try to design the acoustic structure in
order to avoid high strain rates of change within the lossy material layers.

7.1 Experimental Determination of the Reflector Performance

In Papers 2 and 3, the reflector leakage was studied in detail. Marksteiner et al.
found out in [36] that a significant improvement in Q-values of SMR devices can
be obtained by providing shear reflectivity simultaneously with the primary lon-
gitudinal reflectivity – something not achieved by a standard quarter-wavelength
reflector. Resonators with such a shear optimized reflector fabricated by Infineon
Technologies GmbH showing a Q-value approaching 2000 at 1850 MHz were
post-processed so that the substrate was removed under the reflector. For the re-
moval of the substrate, the Si substrate was first thinned and polished to facilitate
performing lithography on the backside of the wafers. Then the substrate was
removed under some, but not all of the resonators. Resonators where electri-
cally characterized both before and after the substrate release leading to having

80
 Figure 55: Experiment performed in Paper 2. a) shows the unpro-
cessed resonators which have roughened backside of the wafer, b)
shows the resonators that have polished and thinned Si substrate under
the reflector, and c) shows the resonators that have no substrate under
the reflector at all.

measurement of three different kind of resonators to compare (Fig. 55): a) un-

touched resonators with thick and roughened substrate underneath, b) resonators
on thinned and polished Si substrate, and c) suspended resonators with no sub-
strate at all.
It was found out that in the resonators with substrate remaining underneath
the reflector but having been polished on the backside, a set of new spurious res-
onances with regular frequency spacing arose in the electrical response (Fig. 56
’after’). This frequency spacing matches that of standing wave resonances in the
substrate. In the resonators where the substrate was removed, these backside re-
flections disappear.
In the Q-values of the resonators, no significant change could be confirmed as
a result of the rework (see Table 4). It was thus concluded that while acoustic en-
ergy clearly leaks to the substrate through the reflector as shown by the existence
of the backside reflections, it is not the limiting factor to Q-value at these devices.
Some other mechanisms of loss are to be dealt with before further optimization
of the reflector can yield improvement in the Q-value. A likely candidate for a
Q-limiting mechanism are the laterally leaking waves, although others must be
considered too [22, 6].
The approach of Marksteiner et al. in [36] to model the effect of having the
acoustic reflector to provide both longitudinal and shear reflectivity is based on
separately calculating longitudinal and shear transmittances of the reflector in 1-
D, and then assuming a constant fraction of the energy to be in the shear waves.
Thus they calculate a composite Q-value combining losses due to longitudinal
and shear wave leakages, that is to be maximized with minimal trade off with
effective coupling coefficient. While clearly very successful in practice, this pic-
ture is a very rough oversimplification of the actual situation. To study the matter
in more depth, an analysis of the reflector transmission characteristics by laser

81
 0.98
before
0.97 after
released
0.96
|S |
11

0.95

0.94

0.93
11.5 MHz

0.92
1830 1840 1850 1860 1870 1880 1890 1900
f (MHz)

Figure 56: Magnitude of S11 measured from resonators before any

rework (’before’), with substrate thinned and polished (’after’), and
substrate removed (’released’). The vertical lines indicate average fs
and fp values. The 11.5 MHz spacing of the dips in the ’after’ -data
correspond to backside reflections in the substrate. This individual
resonator shows increased |S11 | after releasing but the effect is not
systematically present in all devices.
Table 4: Q-value data over all measured devices. B = before process-
ing, A = after processing, R = released resonators. Three methods of
extracting the Q-values were compared: Qp is a direct evaluation of
Eq. (12), Qmech. is Eq. (13) resulting from fit procedure described in
Section 2.3, and QBode,MAX is calculated according to method of Feld
et al. [78]. For more details, see Paper 2.

Condition Range Spread

Qp B 950-1370 420
A 800-1280 480
R 1080-1340 260
Qmech. B 1770-1910 140
A 1800-2010 210
R 2050-2200 150
QBode,MAX B 1530-1840 310
A 1500-1700 200
R 1590-1970 280

82
 40
1100 TE1
1000 30
Frequency (MHz)

900
20

800
TS2
10
700

600 0
TS1
500
−10

400
−20
30 20 10 0 -10 0 0.5 1 1.5 2 2.5 3 (dB)
|β||| = 0 (dB) |β||| (1/µm)

Figure 57: Measured reflector attenuation (73) (right) and 1-D sim-
ulation of reflector attenuation (red line) compared to that extracted
from the measurement (black noisy line) along β|| = 0 .

interferometric measurements of vibration amplitude both on the surface of the

resonator and at the reflector-substrate interface was performed in Paper 3. This
was allowed by a transparent glass substrate used in the samples. The measure-
ment results were compared with predictions of the 1-D transfer matrix model
and the 2-D FEM model. Utilizing Fourier techniques, the reflector transmittance
characteristics were determined as functions of both frequency and lateral wave
number over a wide range.
The measured reflector attenuation shown in Fig. 57 is defined by
Dtop (β|| , f )
Mattn (β|| , f ) = 20 log10 , (73)
Dbot (β|| , f )

where Dtop (β|| , f ) and Dbot (β|| , f ) are the Fourier transformations of the out of
plane vibration amplitude measured at the top and bottom of the device, respec-
tively, and β|| is the circularly averaged wavenumber component along the surface
of the resonator. The line plots on the left of the attenuation, compare the 1-D
simulated reflector attenuation (smooth line) defined by
utop (f )
Mattn (f ) = 20 log10 , (74)
ubot (f )
where utop (f ) and ubot (f ) are the displacements extracted from the top and bot-
tom of the 1-D layer stack as response to constant voltage drive, and the measured
attenuation along the β|| = 0 axis (i.e. at the long lateral wavelength limit). These
data constitute a direct measurement of the reflector performance. It was found
out that in the β|| = 0 limit, the longitudinal 1-D model predicts the behavior of

83
the reflector very well. The simulated and measured attenuations agree within 2
dB in magnitude and the shapes of the frequency responses match closely.
Interesting features are observed in the (β|| , f )-domain, including a strong at-
tenuation maximum on the TE1 curve not coinciding with the operation point of
the resonator and sudden notches in the attenuation on some dispersion curves.
The 2-D FEM modeling was able to reproduce many of the features observed
in (β|| , f )-domain measurements, but it is evident that for quantitatively accurate
predictions one must employ absorbing boundary conditions which work for all
wave modes simultaneously at the lateral and bottom edges of the model. That
provided, 2-D FEM would form a powerful tool for Q-value optimization of the
BAW devices.

84
8 Discussion
Based on high quality AlN piezoelectric thin film and careful design, industrial
state-of-the-art SMR BAW resonators can reach Q-values of around 2600 in the 2
GHz frequency range and effective coupling coefficient K 2 of 6.30 % (using def-
inition (20)) [79]. In this kind of devices, both lateral and vertical energy trapping
must be co-optimized extremely well, the first being influenced by the dispersion
characteristics and the boundary structures of the resonator and the latter by the
longitudinal and shear wave transmission properties of the reflector. The reflector
performance and the dispersion characteristics are both determined by the layer
thicknesses and material properties of the complete layer stack, the control of and
accurate knowledge of which are crucial. State-of-the-art resonators of the air-
gap (FBAR) type not suffering from any kind of acoustic leakage to the substrate,
show Q-values up to 3700 [79] at 2 GHz and effective coupling coefficient K 2
reaching 7 % (definition (20)). In the industrial devices, spurious resonance sup-
pression is based on both boundary frame method and apodization. This work has
concentrated on the investigation and optimization of part of the basic building
blocks that enable realization of high performance BAW resonators.
The state-of-the-art BAW resonators allow for fabrication of high performance
GHz range pass-band filters for the mobile phone radio systems. BAW filters com-
plement the dominating SAW filter technology in the high frequency (>2 GHz)
applications that are most demanding in the insertion loss, pass-band skirt steep-
ness, and temperature stability requirements. Examples of such applications are
the US-PCS duplexer and some WCDMA bands. The AlN BAW ladder filter is
fairly limited in application range due to its pass band width being strictly limited
by the electromechanical coupling coefficient of AlN and due to the lack of an
economical way of providing balanced-unbalanced conversion, which SAW fil-
ters can readily offer. Thus BAW is much less versatile as compared to SAW as a
general filter technology platform. In its present state, the BAW filter technology
does not seem to be on a path to replace a significant portion of the SAW filter
market, but is a necessary component in order to enable fabrication of modern
mobile phones that operate almost anywhere in the world.
The availability of high-Q BAW resonators has raised interest in their usage
in applications other than RF filters. For example, a BAW resonator can serve
as a frequency reference in an oscillator, facilitated by the fact that it is rather
straightforward to fabricate resonators that are temperature stable to the first order
[20, 80, 81]. Thin film BAW resonators coated with a chemically active layer have
also been used as chemical sensors due to the extreme sensitivity for surface mass
loading [82, 83]. No significant commercial success, however, exists yet outside
the field of mobile phone RF filters.
The present 3G and the upcoming 4G/LTE mobile phones will continue to

85
add to the complexity of the radio systems and number of frequency bands at
use. The requirements for the filter performance in bandwidth, pass-band edge
steepness, linearity, and temperature stability are high. A large number of filters
and duplexers need to be fitted into a small footprint and produced in a highly
economical way. A characteristic feature of both SAW and BAW technologies is
their vibrating surface requiring a hermetic package which increases cost and size
of the systems.
For expanding the application range of the thin film BAW technology, the
laterally coupled BAW (LBAW) filter is a promising development. The obtain-
able filter bandwidth may be much larger than in the ladder filters and balanced-
unbalanced functionality seems to be feasible with a reasonable amount of added
complication in the fabrication process.
Another path of further development of the BAW technology would be the
introduction of new low-loss thin film materials exhibiting piezoelectric effect
significantly stronger than that of AlN. Availability of, for example, high quality
industrial level LiNbO3 thin film deposition process, would open up a completely
new range of possibilities for miroacoustics. Recently, Kadota et al. have reported
microacoustic devices based on chemical vapor deposition of LiNbO3 thin films
[84]. A very interesting finding reported in [85] is that ScAlN exhibits up to 400
% increased piezoelectric coefficients as compared to plain AlN.
An interesting development arising from the field of SAW filters is the piezo-
electric boundary wave devices [86]. Filters with attractive performance have
been realized by employing a wave mode concentrated at the interface of a piezo-
electric crystal and another material, where an interdigitated electrode similar to
that used in SAW filters is employed [87]. Perhaps the most remarkable feature of
these devices is that the vibration does not reach the outer surfaces of the device
and thus they do not require a hermetic package like traditional SAW and BAW
components.
In the complicated multi-band mobile phones, it would be highly desirable to
have a software controlled filter, which can adjust to the band or bands (or even
channels within the bands) at use at a given moment. Technology allowing for that
does not exist at the moment – at least with a form factor that could be imagined
to fit into a mobile phone. The frequency of microacoustic filters cannot be effec-
tively tuned, as it would require the change of dimensions or material properties
of solid materials. The latter is possible to some extent in ferroelectric materials
and very strong piezoelectrics, which unfortunately often exhibit large losses. In
principle, a filter could be based on tunable capacitors and inductors, but at GHz-
frequencies their Q-values are low and size fairly large. Digital filtering would
require very high computation power at the GHz range and result in high power
consumption. At the present, it cannot be foreseen if a tunable filter technology
will emerge, and what the technology would be based on, but there certainly will

86
be efforts to realize one in the coming years. Meanwhile, microacoustic filters
consisting of SAW, BAW, boundary wave, and possibly laterally coupled BAW
devices will be the dominant technology for mobile phone RF filters.

87
9 Summary
The performance of a thin film BAW resonator depends most critically on the
quality of the piezoelectric thin film it is based on. Provided a high quality piezo-
material is available, the next design issues of utmost importance are the optimiza-
tion of the energy trapping efficiency in both vertical and lateral directions and the
control of the laterally standing wave resonances. The former is instrumental for
achieving high Q-value resonators and the latter for avoiding spurious ripple in
the resonator and filter responses.
The design, modeling and performance optimization of thin film BAW res-
onators of the SMR type based on acoustic Bragg reflectors were addressed in
this work. The modeling tools used were the 1-D electro-acoustic transfer ma-
trix model for the resonator, a transmission line model for the reflector stack, the
Thomson-Haskell matrix method for plate wave dispersion characteristics, and 2-
D FEM. A dispersion based model for lateral eigenmodes and acoustical intercou-
pling of adjacent resonators allowing for acoustically coupled filter design were
developed. For the characterization of resonators, electrical measurements and
interferometric measurements of vibration fields at the devices, combined with
Fourier techniques to yield dispersion characteristics, were used.
For all modeling, accurate knowledge of the material parameters is crucial.
In order to predict the effects that depend on the dispersion characteristics, it is
necessary to know not only the longitudinal wave parameters but also the shear
wave parameters, which are difficult to measure. In Paper 1, extraction of the full
set of elastic parameters from a thin film material by fitting a dispersion model to
interferometrically measured dispersion curves was demonstrated.
The design of Bragg reflectors of the SMR type devices based on different
materials were discussed in Section 3.2. The industrially successful W − SiO2
reflector, requiring two lithography steps, provides good performance based on
its superior impedance ratio over all-dielectric reflectors and metal-dielectric re-
flectors using metals lighter than W . It was shown, however, that with a hybrid
reflector consisting of several layer pairs of Ta2 O5 and SiO2 and only a single
layer of Pt requiring a single lithography, high performance SMRs can be real-
ized.
In Paper 2, an experiment was conducted to find out whether the Q-value of a
2 GHz SMR having a shear-optimized reflector can still be improved from around
2000, but it was found out that the leakage through reflector, although clearly
present, was not the Q-limiting factor in those devices – other loss mechanisms
have to be sought for further improvement. To study the reflector performance
in more detail, a laser interferometric measurement of vibration fields on top and
at the bottom of an SMR on a transparent substrate was conducted in Paper 3.
The 1-D model was found to agree with the experimental reflector performance at

88
the long lateral wavelength limit very well, but features depending on the lateral
wavelength were discovered. From comparison of measurement and 2-D FEM
modeling results, it was concluded that with improvements in handling of acoustic
leakage at the boundaries, the 2-D FEM could be used as a powerful tool for
advanced reflector optimization.
The most important lateral plate wave dispersion-related effects in BAW res-
onators are the lateral energy trapping and laterally standing plate wave reso-
nances, the spurious resonances. The suppression of the lateral resonances by the
boundary frame method was studied in detail by dispersion based 2-D and 2-D
FEM modeling and a comparison to actual devices in Paper 4, where it was found
out that two suppression effects are present in such structures – a narrow band
single-mode suppression and a wide band suppression, the latter being related to
the approximate orthogonality of the eigenmodes.
The dispersion based modeling of laterally standing waves was further de-
veloped to facilitate simulation of harmonic response of devices consisting of an
arbitrary number of acoustically coupled adjacent resonators forming an electrical
N -port (Paper 5). This allowed for design of experimental laterally coupled BAW
filters (Paper 6, Figs. 49–50), which shows promise as a new type of microacous-
tic filter combining desired features of both SAW and BAW technologies.

89
References
[1] 3rd Generation Partnership Project (3GPP), “Technical Specification 05.05
V8.20.0 (2005-11),” 2005.
[2] Nokia Oyj, http://www.nokia.fi/tuotteet/kaikki-puhelimet/nokia-n8/tarkka-
tuoteseloste, August 2010.
[3] R. Aigner, “SAW and BAW technologies for RF filter applications: A review
of the relative strengths and weaknesses,” in Proc. IEEE Ultrasonics Sympo-
sium, 2008, pp. 582–589.
[4] I. Larson, J.D., R. Ruby, P. Bradley, and Y. Oshmyansky, “A BAW Antenna
Duplexer for the 1900 MHz PCS Band,” in Proc. IEEE Ultrasonics Sympo-
sium, 1999, pp. 887–890.
[5] M. Kadota, T. Nakao, N. Taniguchi, E. Takata, M. Mimura, K. Nishiyama,
T. Hada, and T. Komura, “Surface Acoustic Wave Duplexer for US Personal
Communication Services with Good Temperature Characteristics,” Japanese
Journal of Applied Physics, vol. 44, no. 6B, pp. 4527–4531, 2005.
[6] K.-Y. Hashimoto, Ed., RF Bulk Acoustic Wave Filters for Communications.
Artech House, 2009.
[7] C. Lam, “A review of the recent development of MEMS and crystal oscillators
and their impacts on the frequency control products industry,” in Proc. IEEE
Ultrasonics Symposium, 2008, pp. 694–704.
[8] R. Kinsman, “A history of crystal filters,” in Proc. IEEE Frequency Control
Symposium, 1998, pp. 563–570.
[9] K. Lakin, J. Wang, G. Kline, A. Landin, Y. Chen, and J. Hunt, “Thin Film
Resonators and Filters,” in Proc. IEEE Ultrasonics Symposium, 1982, pp.
466–475.
[10] R. Ruby, “Micromachined cellular filters,” in IEEE MTT-S Microwave Sym-
posium Digest, 1996, pp. 1149–1152.
[11] W. E. Newell, “Face-mounted piezoelectric resonators,” in Proceedings of
the IEEE, vol. 53, no. 6, 1965, pp. 575–581.
[12] P. Bradley, R. Ruby, A. Barfknecht, F. Geefay, C. Han, G. Gan, Y. Osh-
myansky, and I. Larson, J.D., “A 5 mm × 5 mm × 1.37 mm hermetic FBAR
duplexer for PCS handsets with wafer-scale packaging,” in Proc. IEEE Ultra-
sonics Symposium, 2002, pp. 931–934.

90
[13] G. Fattinger, A. Volatier, R. Aigner, and F. Dumont, “BAW PCS-Duplexer
Chipset and Duplexer Applications,” in Proc. IEEE Ultrasonics Symposium,
2008, pp. 602–606.

[14] E. Schmidhammer, H. Heinze, M. Schmiedgen, M. Mayer, and A. Link,

“High Volume Production of a Fully Matched 5050 PCS-CDMA-BAW Du-
plexer,” in Proc. IEEE Ultrasonics Symposium, 2006, pp. 329–332.

[15] K. M. Lakin and J. S. Wang, “Acoustic bulk wave composite resonators,”

Applied Physics Letters, vol. 38, no. 3, pp. 125–127, 1981.

[16] G. Perluzzo, C. Jen, and E. Adler, “Characteristics of reactive magnetron

sputtered ZnO films,” in Proc. IEEE Ultrasonics Symposium, 1989, pp. 373–
376.

[17] C. Vale, J. Rosenbaum, S. Horwitz, S. Krishnaswamy, and R. Moore, “FBAR

filters at GHz frequencies,” 1990, pp. 332–336.

[18] P. Osbond, C. Beck, C. Brierley, M. Cox, S. Marsh, and N. Shorrocks, “The

influence of ZnO and electrode thickness on the performance of thin film bulk
acoustic wave resonators,” in Proc. IEEE Ultrasonics Symposium, 1999, pp.
911–914.

[19] J. Molarius, J. Kaitila, T. Pensala, and M. Ylilammi, “Piezoelectric ZnO

films by r.f. sputtering,” Journal of Materials Science: Materials in Electron-
ics, vol. 14, no. 5, pp. 431–435, 2003.

[20] K. Lakin, K. McCarron, and J. McDonald, “Temperature compensated bulk

acoustic thin film resonators,” in Proc. IEEE Ultrasonics Symposium, 2000,
pp. 855–858.

[21] B. Auld, Acoustic Fields and Waves in Solids, 2nd ed. Robert E. Krieger
Publishing Company, 1990, vol. I.

[22] R. Ruby, “Review and Comparison of Bulk Acoustic Wave FBAR, SMR
Technology,” in Proc. IEEE Ultrasonics Symposium, 2007, pp. 1029–1040.

[23] J. Kaitila, M. Ylilammi, J. Ella, and R. Aigner, “Spurious resonance free

bulk acoustic wave resonators,” in Proc. IEEE Ultrasonics Symposium, 2003,
pp. 84–87.

[24] J. Rosenbaum, Bulk Acoustic Wave Theory and Devices. Artech House,
1988.

91
[25] J. Larson, P. Bradley, S. Wartenberg, and R. Ruby, “Modified Butterworth-
Van Dyke circuit for FBAR resonators and automated measurement system,”
in Proc. IEEE Ultrasonics Symposium, 2000, pp. 863–868.

[26] IEEE Standard on Piezoelectricity, Std 176-1987. ANSI/IEEE, 1987, p. 51.

[27] K. Yoo, H. Uberall, and J. Williams, W., “Spurious Resonances in Bulk

Acoustic Wave Resonators,” in Proc. IEEE Ultrasonics Symposium, 1982,
pp. 490–493.

[28] B. Auld, Acoustic Fields and Waves in Solids, 2nd ed. Robert E. Krieger
Publishing Company, 1990, vol. II.

[29] R. Aigner, “Bringing BAW technology into volume production: The ten
commandments and the seven deadly sins,” in Proc. 3rd International Sympo-
sium on Acoustic Wave Devices for Future Mobile Communication Systems,
2007.

[30] T. Pensala, M. Ylilammi, J. Meltaus, and K. Kokkonen, “Area and Disper-

sion Dependence of Vibration Shape and Coupling Coefficient in Thin Film
BAW Resonators,” in Proc. IEEE Ultrasonics Symposium, 2007, pp. 1661–
1664.

[31] S. Chang, N. Rogacheva, and C. Chou, “Analysis of methods for determin-

ing electromechanical coupling coefficients of piezoelectric elements,” IEEE
Trans. on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 42, no. 4,
pp. 630–640, 1995.

[32] J. Kaitila, “Review of Wave Propagation in BAW Thin Film Devices –

Progress and Prospects,” in Proc. IEEE Ultrasonics Symposium, 2007, pp.
120–129.

[33] W. P. Mason, Piezoelectric Crystals and Their Applications to Ultrasonics.

D. Van Nostrand, 1950.

[34] M. Ylilammi, “Handbook of Bulk Acoustic Wave Resonators and Filters,”

2000, VTT Internal report.

[35] E. Voipio, Siirtojohtojen teoria. Otakustantamo, 1973.

[36] S. Marksteiner, J. Kaitila, G. Fattinger, and R. Aigner, “Optimization of

Acoustic Mirrors for Solidly Mounted BAW Resonators,” in Proc. IEEE Ul-
trasonics Symposium, 2005, pp. 329–332.

92
[37] A. Jansman, F. Vanhelmont, R. Milsom, J. Ruigrok, F. de Bruijn, R. Lufting,
and J.-W. Lobeek, “Bulk acoustic wave resonators and filters – 2D modelling
and industrialization aspects,” in Proc. IEEE Ultrasonics Symposium, vol. 1,
2005, pp. 97–100.

[38] J.-W. Lobeek, R. Strijbos, A. Jansman, N. X. Li, A. Smolders, and N. Puls-

ford, “High-Q BAW Resonator on Pt/Ta2O5/SiO2-based Reflector Stack,” in
IEEE MTT-S Microwave Symposium Digest, 2007, pp. 2047–2050.

[39] R. Milsom, “Two-Dimensional Theory of Thin-Film ZnO Resonators on Sil-

icon,” in Proc. IEEE Ultrasonics Symposium, 1982, pp. 484–489.

[40] K. Lakin, “Numerical analysis of two dimensional thin film resonators,”

Proc. IEEE International Frequency Control Symposium, pp. 502–508, 1993.

[41] R. Ruby, J. Larson, R. Fazzio, and C. Feng, “Performance degradation ef-

fects in FBAR filters and resonators due to lamb wave modes,” in Proc. IEEE
Ultrasonics Symposium, 2005, pp. 1832–1835.

[42] M. Lowe, “Matrix Techniques for Modeling Ultrasonic Waves in Multilay-

ered Media,” IEEE Trans. on Ultrasonics, Ferroelectrics, and Frequency Con-
trol, vol. 42, no. 4, pp. 525–541, 1995.

[43] R. Milsom, H.-P. Lobl, C. Metzmacher, P. Lok, A. Tuinhout, and F. van

Straten, “2D model of solidly-mounted and membrane BAW devices,” in
Proc. IEEE Ultrasonics Symposium, 2003, pp. 1802–1807.

[44] W. Shockley, D. Curran, and D. Koneval, “Energy Trapping and Related

Studies of Multiple Electrode Filter Crystals,” in Proc. 17th Annual Sympo-
sium on Frequency Control, 1963, pp. 88–126.

[45] J. Meltaus, T. Pensala, and K. Kokkonen, “Modelling of 2-D lateral modes

in solidly-mounted BAW resonators,” in Proc. IEEE Ultrasonics Symposium,
2008, pp. 1544–1547.

[46] T. Makkonen, A. Holappa, J. Ella, and M. Salomaa, “Finite element simula-

tions of thin-film composite BAW resonators ,” IEEE Trans. on Ultrasonics,
Ferroelectrics, and Frequency Control, vol. 48, no. 5, pp. 1241–1258, 2001.

[47] R. Thalhammer, G. Fattinger, M. Handtmann, and S. Marksteiner, “Ohmic

effects in BAW-resonators,” in IEEE MTT-S Microwave Symposium Digest,
2006, pp. 390–393.

93
[48] R. Thalhammer, J. Kaitila, R. Aigner, and S. Marksteiner, “Prediction of
BAW resonator performance using experimental and numerical methods,” in
Proc. IEEE Ultrasonics Symposium, 2004, pp. 282–285.

[49] M. Mayer, S. Zaglmayr, K. Wagner, and J. Schoberl, “Perfectly Matched

Layer Finite Element Simulation of Parasitic Acoustic Wave Radiation in Mi-
croacoustic Devices,” in Proc. IEEE Ultrasonics Symposium, 2007, pp. 702–
706.

[50] D. Ekeom, B. Dubus, and A. Volatier, “Solidly Mounted Resonator (SMR)

FEM-BEM Simulation,” in Proc. IEEE Ultrasonics Symposium, 2006, pp.
1474–1477.

[51] J. D. L. III, R. C. Ruby, and P. Bradley, U.S. Patent 6 215 375 B1, Apr. 10
2001.

[52] D. Rosen, J. Bjurstrom, and I. Katardjiev, “Suppression of spurious lateral

modes in thickness-excited FBAR resonators,” IEEE Trans on. Ultrasonics,
Ferroelectrics and Frequency Control, vol. 52, no. 7, pp. 1189–1192, 2005.

[53] A. Link, E. Schmidhammer, H. Heinze, M. Mayer, B. Bader, and R. Weigel,

“Appropriate Methods to Suppress Spurious FBAR Modes in Volume Produc-
tion,” in IEEE MTT-S Microwave Symposium Digest, 2006, pp. 394–397.

[54] T. Pensala, M. Ylilammi, J. Meltaus, and K. Kokkonen, “Area and Disper-

sion Dependence of Vibration Shape and Coupling Coefficient in Thin Film
BAW Resonators,” in Proc. IEEE Ultrasonics Symposium, 2007, pp. 1661–
1664.

[55] G. Fattinger, S. Marksteiner, J. Kaitila, and R. Aigner, “Optimization of

acoustic dispersion for high performance thin film BAW resonators,” in Proc.
IEEE Ultrasonics Symposium, vol. 2, 2005, pp. 1175–1178.

[56] R. Sykes and W. Beaver, “High Frequency Monolithlc Crystal Filters with
Posslble Application to Single Frequency and Single Side Band Use,” in Proc.
20th Annual Symposium on Frequency Control, 1966, pp. 288–308.

[57] K. Lakin, J. Belsick, J. McDonald, and K. McCarron, “High performance

stacked crystal filters for GPS and wide bandwidth applications,” in Proc.
IEEE Ultrasonics Symposium, 2001, pp. 833–838.

[58] K. Lakin, “Coupled resonator filters,” in Proc. IEEE Ultrasonics Symposium,

2002, pp. 901–908.

94
[59] G. Fattinger, J. Kaitila, R. Aigner, and W. Nessler, “Single-to-balanced filters
for mobile phones using coupled resonator BAW technology,” in Proc. IEEE
Ultrasonics Symposium, 2004, pp. 416–419.

[60] A. Volatier, E. Defay, A. N’hari, J. Carpentier, P. Ancey, and B. Dubus,

“Design, Elaboration and Characterization of Coupled Resonator Filters for
WCDMA Applications,” in Proc. IEEE Ultrasonics Symposium, 2006, pp.
829–832.

[61] M. Small, T. Jamneala, L. Callaghan, J. Larson, and R. Ruby, “A De-

Coupled Stacked Bulk Acoustic Resonator (DSBAR) Filter With 2 dB Band-
width > 4% ,” in Proc. IEEE Ultrasonics Symposium, 2007, pp. 604–607.

[62] T. Grudkowski, J. Black, G. Drake, and D. Cullen, “Progress in the Develop-

ment of Miniature Thin Film Baw Resonator and Filter Technology,” in Proc.
36th Annual Symposium on Frequency Control, 1982, pp. 537–548.

[63] V. Yantchev and I. Katardjiev, “Micromachined thin film plate acoustic res-
onators utilizing the lowest order symmetric Lamb wave mode,” IEEE Trans.
on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 54, no. 1, pp. 87–
95, 2007.

[64] W. Pan, R. Abdolvand, and F. Ayazi, “A low-loss 1.8GHz monolithic thin-

film piezoelectric-on-substrate filter,” in Proc. IEEE 21st International Con-
ference on Micro Electro Mechanical Systems, 2008, pp. 176–179.

[65] J. Meltaus, T. Pensala, K. Kokkonen, and A. Jansman, “Laterally coupled

solidly mounted BAW resonators at 1.9 GHz,” in Proc. IEEE Ultrasonics
Symposium, 2009, pp. 847–850.

[66] J. E. Graebner, B. P. Barber, P. L. Gammel, D. S. Greywall, and S. Gopani,

“Dynamic visualization of subangstrom high-frequency surface vibrations,”
Applied Physics Letters, vol. 78, no. 2, pp. 159–161, 2001.

[67] G. G. Fattinger and P. T. Tikka, “Modified Mach-Zender laser interferometer

for probing bulk acoustic waves,” Applied Physics Letters, vol. 79, no. 3, pp.
290–292, 2001.

[68] H. Martinussen, A. Aksnes, and H. E. Engan, “Wide frequency range mea-

surements of absolute phase and amplitude ofvibrations in micro- and nanos-
tructures by optical interferometry,” Opt. Express, vol. 15, no. 18, pp. 11 370–
11 384, 2007.

95
[69] K.-Y. Hashimoto, K. Kashiwa, T. Omori, M. Yamaguchi, O. Takano, S. Me-
guro, and K. Akahane, “A fast scanning laser probe based on Sagnac inter-
ferometer for RF surface and bulk acoustic wave devices,” in IEEE MTT-S
Microwave Symposium Digest, 2008, pp. 851–854.
[70] H. Yatsuda, S. Kamiseki, and T. Chiba, “Calculation and optical measure-
ment of SAW diffraction pattern of slanted finger SAW filters on YZ LiNbO3
and 128YX LiNbO3 ,” in Proc. IEEE Ultrasonics Symposium, 2001.
[71] J. V. Knuuttila, P. T. Tikka, and M. M. Salomaa, “Scanning Michelson inter-
ferometer for imaging surface acoustic wave fields,” Opt. Lett., vol. 25, no. 9,
pp. 613–615, 2000.
[72] K. Kokkonen and M. Kaivola, “Scanning heterodyne laser interferometer
for phase-sensitive absolute-amplitude measurements of surface vibrations,”
Applied Physics Letters, vol. 92, no. 6, 2008.
[73] J. Graebner, H. Safar, B. Barber, P. Gammel, G. Herbsommer, L. Fetter,
J. Pastalan, H. Huggins, and R. Miller, “Optical mapping of surface vibrations
on a thin-film resonator near 2 GHz,” in Proc. IEEE Ultrasonics Symposium,
2000, pp. 635–638.
[74] K. Kokkonen and T. Pensala, “Laser Interferometric Measurements and Sim-
ulations of Waves Transmitted Through the Acoustic Mirror in Thin Film
BAW Resonator,” in Proc. IEEE Ultrasonics Symposium, 2006, pp. 460–463.
[75] A. Link, E. Schmidhammer, M. Mayer, B. Bader, K. Wagner, and R. Weigel,
“2D-FEM Modeling of Mirror-Type FBARs Based on an Adapted Set of Ma-
terial Parameters,” in Proc. IEEE Ultrasonics Symposium, 2006, pp. 452–455.
[76] A. Hagelauer, B. Bader, G. Henn, A. Schaeufele, S. Marksteiner, K. Wagner,
and R. Weigel, “Low-loss BAW filters on high resistivity silicon for mobile
radio,” in IEEE MTT-S Microwave Symposium Digest, 2009, pp. 337–340.
[77] K. Lakin, J. Belsick, J. McDonald, and K. McCarron, “Improved bulk wave
resonator coupling coefficient for wide bandwidth filters,” in Proc. IEEE Ul-
trasonics Symposium, 2001, pp. 827–831.
[78] D. Feld, R. Parker, R. Ruby, P. Bradley, and S. Dong, “After 60 years: A new
formula for computing quality factor is warranted,” in Proc. IEEE Ultrasonics
Symposium, 2008, pp. 431–436.
[79] R. Ruby, R. Parker, and D. Feld, “Method of Extracting Unloaded Q Ap-
plied Across Different Resonator Technologies,” in Proc. IEEE Ultrasonics
Symposium, 2008, pp. 1815–1818.

96
[80] M. Allah, J. Kaitila, R. Thalhammer, W. Weber, and D. Schmitt-Landsiedel,
“Temperature compensated solidly mounted bulk acoustic wave resonators
with optimum piezoelectric coupling coefficient,” in proc. IEEE Electron De-
vices Meeting, 2009, pp. 1–4.

[81] Q. Zou, D. Lee, F. Bi, R. Ruby, M. Small, S. Ortiz, Y. Oshmyansky, and

J. Kaitila, “High coupling coefficient Temperature compensated FBAR res-
onator for oscillator application with wide pulling range,” in Proc IEEE Fre-
quency Control Symposium, 2010, pp. 646–651.

[82] R. Gabl, E. Green, M. Schreiter, H. Feucht, H. Zeininger, R. Primig,

D. Pitzer, G. Eckstein, and W. Wersing, “Novel integrated FBAR sensors:
a universal technology platform for bio- and gas-detection,” in Proc. IEEE
Conference on Sensors, vol. 2, 2003, pp. 1184–1188.

[83] J. Enlund, D. Martin, V. Yantchev, and I. Katardjiev, “FBAR sensor array for
in liquid operation,” IEEE Sensors Journal, no. 99, p. 1, 2010.

[84] M. Kadota, T. Ogami, K. Yamamoto, Y. Negoro, and H. Tochishita, “High

frequency Lamb wave device composed of LiNbO3 thin film,” in Proc. IEEE
Ultrasonics Symposium, 2008, pp. 1940–1943.

[85] M. Akiyama, T. Kamohara, K. Kano, A. Teshigahara, Y. Takeuchi, and

N. Kawahara, “Enhancement of Piezoelectric Response in Scandium Alu-
minum Nitride Alloy Thin Films Prepared by Dual Reactive Cosputtering,”
Advanced Materials, vol. 21, pp. 593–596, 2009.

[86] K.-Y. Hashimoto, Y. Wang, T. Omori, M. Yamaguchi, M. Kadota, H. Kando,

and T. Shibahara, “Piezoelectric boundary acoustic waves: Their underlying
physics and applications,” in Proc. IEEE Ultrasonics Symposium, 2008, pp.
999–1005.

[87] H. Kando, D. Yamamoto, M. Mimura, T. Oda, A. Shimizu, K. Shimoda,

E. Takata, T. Fuyutsume, R. Kubo, and M. Kadota, “RF Filter using Boundary
Acoustic Wave,” in Proc. IEEE Ultrasonics Symposium, 2006, pp. 188–191.

97
 Papers 1–6 are not included in the PDF version.
Please order the printed version to get the complete publication
(http://www.vtt.fi/publications/index.jsp).

7
 Series title, number and
report code of publication

VTT Publications 756

VTT-PUBS-756
Author(s)
Tuomas Pensala
Title

Thin Film Bulk Acoustic Wave Devices

Performance Optimization and Modeling
Abstract
Thin film bulk acoustic wave (BAW) resonators and filters operating in the GHz range are
used in mobile phones for the most demanding filtering applications and complement the
surface acoustic wave (SAW) based filters. Their main advantages are small size and
high performance at frequencies above 2 GHz. This work concentrates on the characteri-
zation, performance optimization, and modelling techniques of thin film BAW devices.
Laser interferometric vibration measurements together with plate wave dispersion
modeling are used to extract the full set of elastic material parameters for sputter depos-
ited ZnO, demonstrating a method for obtaining material data needed for accurate simula-
tion of the devices.
The effectiveness of the acoustic interference reflector used to isolate the vibration
from the substrate is studied by 1-D modeling, 2-D finite element method and by electrical
and laser interferometric measurements. It is found that the Q-value of reflector-based
BAW resonators operating at 2 GHz is limited to approximately 2000 by mechanisms
other than leakage through the reflector.
Suppression of spurious resonances in ZnO resonators is studied in depth by modeling
and measurements. It is verified that the approximate mode orthogonality is behind the sup-
pression in boundary frame type ZnO devices operating in the piston mode, but also another
narrow band mode suppression mechanism is found.
A plate wave dispersion based 2-D simulation scheme for laterally acoustically coupled
BAW resonator filters is developed and employed in designing of experimental devices,
which show both good agreement with the model predictions and a remarkable 4.9 %
relative bandwidth.

ISBN
978-951-38-7722-4 (soft back ed.)
978-951-38-7723-1 (URL: http://www.vtt.fi/publications/index.jsp)
Series title and ISSN Project number
VTT Publications 70339
1235-0621 (soft back ed.)
1455-0849 (URL: http://www.vtt.fi/publications/index.jsp)
Date Language Pages
January 2011 English, Finnish abstr. 97 p. + app. 73 p.
Name of project Commissioned by

Keywords Publisher
Thin film, bulk acoustic wave, BAW, AlN, ZnO, VTT Technical Research Centre of Finland
resonator, filter, GHz, microacoustic, modeling, P.O. Box 1000, FI-02044 VTT, Finland
interferometric imaging Phone internat. +358 20 722 4520
Fax +358 20 722 4374
 Julkaisun sarja, numero ja
raporttikoodi

VTT Publications 756

VTT-PUBS-756
Tekijä(t)
Tuomas Pensala
Nimeke
Pietsosähköiset ohutkalvoresonaattorit
Suorituskyvyn optimointi ja mallinnus
Tiivistelmä
Pietsosähköisiin ohutkalvoihin perustuvia GHz-alueen mikroakustisia tilavuusaaltoreso-
naattoreita (BAW) ja niihin pohjautuvia suodattimia käytetään matkapuhelinteknologiassa
kaikkein vaativimmissa suodatuskohteissa. Teknologian keskeiset edut ovat pieni koko ja
hyvä suorituskyky korkeilla ( 2 GHz) taajuuksilla. Työssä keskityttiin BAW-resonaattoreiden
karakterisointiin, suorituskyvyn optimointiin ja mallinnustekniikkoihin.
Laserinterferometrimittauksia ja Lambin aaltojen dispersion mallinnusta käytetään yh-
dessä ohutkalvomateriaalin elastisten materiaaliparametrien määrittämiseen. Näin de-
monstroidaan menetelmä, jolla voidaan parantaa mm. mallinnuksen tarkkuutta paremman
materiaaliparametritiedon kautta. Resonaattorit substraatista erottavan akustisen peilin
ominaisuuksia analysoidaan 1D-malleilla, 2D-elementtimenetelmällä sekä kokeellisesti
sähköisin ja laserinterferometrimittauksin. Akustista peiliä parantamalla n. 2 GHz taajuu-
della toimivien resonaattorien hyvyyslukua ei saada parannettua juurikaan yli 2000:n.
Sivuresonanssien vaimennusta reunapaksunnosmenetelmällä ZnO-resonaattoreissa
analysoidaan mallinnuksella ja mittauksin. Tulokset osoittavat, että sivuresonanssit saa-
daan vaimennettua laajalla kaistalla niiden ominaismuotojen heikkoon lineaariseen riippu-
vuuteen perustuen, mutta myös toinen kapeakaistainen vaimennusefekti havaitaan.
Laatta-aaltojen dispersio-ominaisuuksiin perustuen kehitetään tehokas 2D-mallinnus-
menetelmä, jolla voidaan simuloida horisontaalisesti akustisesti kytkettyjä suodattimia.
Mallia vertaillaan kokeellisiin suodattimiin hyvin tuloksin, ja sen avulla suunniteltu kokeel-
linen suodatin saavuttaa 4,9 % suhteellisen kaistanleveyden 2 GHz:n taajuudella.

ISBN
978-951-38-7722-4 (nid.)
978-951-38-7723-1 (URL: http://www.vtt.fi/publications/index.jsp)
Avainnimeke ja ISSN Projektinumero
VTT Publications 70339
1235-0621 (nid.)
1455-0849 (URL: http://www.vtt.fi/publications/index.jsp)
Julkaisuaika Kieli Sivuja
Tammikuu 2011 Englanti, suom. tiiv. 97 s. + liitt. 73 s.
Projektin nimi Toimeksiantaja(t)

Avainsanat Julkaisija
Thin film, bulk acoustic wave, BAW, AlN, ZnO, VTT
resonator, filter, GHz, microacoustic, modeling, PL 1000, 02044 VTT
interferometric imaging Puh. 020 722 4520
Faksi 020 722 4374
VTT CREATES BUSINESS FROM TECHNOLOGY
Technology and market foresight • Strategic research • Product and service development • IPR and licensing
• Assessments, testing, inspection, certification • Technology and innovation management • Technology partnership
Dissertation VTT PUBLICATIONS 756

• • • VTT PUBLICATIONS 756

VTT PUBLICATIONS
738 Fusion Yearbook. Association EURATOM-TEKES. Annual Report 2009. Eds. by Seppo
Karttunen & Markus Airila. 2010. 136 p. + app. 13 p.
739 Satu Hilditch. Identification of the fungal catabolic D-galacturonate pathway. 2010.
74 p. + app. 38 p.
740 Mikko Pihlatie. Stability of Ni-YSZ composites for solid oxide fuel cells during reduction
and re-oxidation. 2010. 92 p. + app. 62 p.
741 Laxmana Rao Yetukuri. Bioinformatics approaches for the analysis of lipidomics

THIN FILM BULK ACOUSTIC WAVE DEVICES. PERFORMANCE OPTIMIZATION AND MODELING
data. 2010. 75 p. + app. 106 p.
742 Elina Mattila. Design and evaluation of a mobile phone diary for personal health
management. 2010. 83 p. + app. 48 p.
743 Jaakko Paasi & Pasi Valkokari (eds.). Elucidating the fuzzy front end – Experiences
from the INNORISK project. 2010. 161 p.
744 Marja Vilkman. Structural investigations and processing of electronically and protonically
conducting polymers. 2010. 62 p. + app. 27 p.
745 Juuso Olkkonen. Finite difference time domain studies on sub-wavelength aperture
structures. 2010. 76 p. + app. 52 p.
746 Jarkko Kuusijärvi. Interactive visualization of quality Variability at run-time. 2010.
111 p.
747 Eija Rintala. Effects of oxygen provision on the physiology of baker’s yeast Saccharomyces
cerevisiae. 2010. 82 p. + app. 93 p.
748 Virve Vidgren. Maltose and maltotriose transport into ale and lager brewer’s yeast
strains. 2010. 93 p. + app. 65 p.
749 Toni Ahonen, Markku Reunanen & Ville Ojanen (eds.). Customer value driven
service business development. Outcomes from the Fleet Asset Management Project.
2010. 43 p. + app. 92 p.
750 Tiina Apilo. A model for corporate renewal. Requirements for innovation management.
2010. 167 p. + app. 16 p.
751 Sakari Stenudd. Using machine learning in the adaptive control of a smart environment.
2010. 75 p. Tuomas Pensala
752 Evanthia Monogioudi. Enzymatic Cross-linking of ββ-casein and its impact on digestibility
and allergenicity. 2010. 85 p. + app. 66 p.
753 Jukka-Tapani Mäkinen. Concurrent engineering approach to plastic optics design.
2010. 99 p. + app. 98 p.
Thin Film Bulk Acoustic Wave
754 Sanni Voutilainen. Fungal thermostable cellobiohydrolases. Characterization and
protein engineering studies. 2010. 98 p. + app. 55 p.
Devices
756 Tuomas Pensala. Thin Film Bulk Acoustic Wave Devices. Performance Optimization
and Modeling. 2011. 97 p. + app. 73 p. Performance Optimization and Modeling

ISBN 978-951-38-7722-4 (soft back ed.) ISBN 978-951-38-7723-1 (URL: http://www.vtt.fi/publications/index.jsp)

ISSN 1235-0621 (soft back ed.) ISSN 1455-0849 (URL: http://www.vtt.fi/publications/index.jsp)